# Responsive MCP: research the client before you write the proposal

> After the Responsive MCP fall release, we let an AI agent check a client's last 30 days. 17 calls on three companies cost $0.0195; one brief costs $0.0068.

- Canonical page: https://everyinfra.com/en/blog/responsive-mcp-sales-proposal-account-research
- Published: 2026-11-16T01:45:33+08:00 · Fact-checked: 2026-11-16T01:10:00+08:00

We ran a small test on two listed US companies, Wayfair and Snowflake. The goal was simple: collect what a proposal writer needs before writing the "our understanding of your business" section. On November 16, 2026, between 00:18 and 00:34 Singapore time, 17 charged calls returned company pages, company posts, open roles, news and web results. The total charge was $0.019516. One complete brief for one client takes 6 calls and costs $0.006806. At 100 proposals a year, the data cost is about $0.68.

Four of those calls were for a third company, Mercado Libre. We pulled its open roles through a remote MCP `tools/call`. That call shows what the same step looks like inside an AI assistant. Every number in this article comes from the raw responses of these calls. A script recomputed each count from those files.

Why does this matter for Responsive users? On November 10, 2026, Responsive shipped its fall release. The release moves MCP from "find an answer" to "do the work" inside ChatGPT, Microsoft Copilot and Claude. A seller can answer a security questionnaire and send a follow-up in one conversation. A security expert can draft and save assigned answers. A proposal team can capture new knowledge, tag it and send it for review.

A separate Responsive help article covers the other direction. Responsive can also act as an MCP client, so an admin can connect an outside MCP server to Responsive AI. We have no Responsive account to test with. Every Responsive, Claude, ChatGPT and Copilot step below is a configuration example. Each one is based on the vendor's own help pages.

Each company returned 20 posts, but the spread was very different. All 20 Snowflake posts fell in the last 30 days, while only 1 of 20 Mercado Libre posts did. A job search on the word "Snowflake" returned 50 roles, and only 12 were Snowflake's own. The company-filter URL from its profile returned 50 out of 50. Three of Wayfair's six "sales" roles were retail store managers with "Sales" in the title. They have no link to business buying.

> **Key points**
>
> - On November 10, 2026, Responsive extended MCP so ChatGPT, Copilot and Claude can draft and save answers, add comments and complete reviews. Each user keeps their Responsive permissions. Responsive can also act as an MCP client.
> - One client brief uses one call each for company page, posts, open roles, one job detail, news and web search: $0.006806. Our three-company test used 17 charged calls for $0.019516.
> - Sort company posts by `posted_at` and cut the 30-day window before the model sees them. Each company returned 20 posts: 12, 20 and 1 fell in the last 30 days for Wayfair, Snowflake and Mercado Libre.
> - Two hiring-signal mistakes are easy to make. A company name that is also a skill pulls in other employers. Store roles with "Sales" in the title look like sales growth.
> - Keep public data and client-supplied material in two columns. Give every sentence a source ID. Open the original page before you quote a web result.

![Illustration: four information cards on the left show a company building, a picture, a briefcase and a thumbs-up. Dotted lines carry them into a server with a teal light in the middle. The server links to an open proposal document on a desk at the right, with a teal block at the top. A gray folder with a padlock sits apart next to the document.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/img-explain.webp)

Account research before a proposal: public pages, posts, roles and news flow through one MCP server into one section of the proposal. Client material stays in a separate, locked folder.

## The job: read the client's last month before you answer the RFP

Most RFP responses and sales proposals have a section about the client. It may be called "our understanding of your situation" or "project background". Evaluators read it to see if the bidder made an effort to understand them. The US Federal Acquisition Regulation, FAR 15.305, says proposal evaluation assesses the proposal and the offeror's ability to perform the contract. Agencies must score proposals only on the factors in the solicitation. Private buyers are less formal, but the logic is the same.

A vague background section makes even a strong proposal feel like a template. Most of the raw material for this section is public. It includes the client's size and headquarters, its posts from the last month, its open roles, and its news and earnings. A presales engineer who does this by hand moves between the company page, the careers page, news sites and the investor relations site.

Manual research also misses dates. In this test, the 10 web results for Wayfair included a layoffs story from August 2022. An AI assistant can do this step, but only under two conditions. It needs structured data with dates. It also needs a source for every item, so a person can check it.

Responsive calls its category Strategic Response Management. It manages RFPs, security questionnaires and due diligence questionnaires. The Responsive MCP server brings approved internal answers into AI assistants. Client news is outside data, so it needs a second source. Together, the two sources give a sales proposal AI agent both halves of the job: approved internal answers and fresh external facts.

This test covers only the external half. Our pass rule was two real companies, at least four kinds of public data per company, and one remote MCP round trip. We chose Wayfair and Snowflake because one is a retailer and one is a software company. Their hiring patterns differ a lot, which makes hiring signals easier to compare. Mercado Libre adds a Latin American sample.

![Animated flow. Four source boxes on the left: company page company_profile, company posts company_posts, open roles jobs, and news google_news / web. Arrows meet at the first stop, "EveryInfra remote MCP, everyinfra_call_api". The second stop is "Claude, ChatGPT or Copilot: sort, code, cite". The third stop is "Responsive project section: Our understanding", with a Review box below it. A teal dot travels from the sources through the three stops to Review every eight seconds.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-01-brief-flow.en.light.svg)

## What Responsive opened on November 10

Responsive published the fall release news on November 10, 2026. The first paragraph says Responsive "today announced expanded Model Context Protocol (MCP) capabilities". Teams can now use ChatGPT, Microsoft Copilot and Claude to reach Responsive knowledge, collaborate and move response work forward. The first MCP release came with the spring release on May 13, 2026. That release focused on answering client questions and generating emails and proposals from approved content.

The fall release puts the focus on actions. The [fall release news](https://www.responsive.io/news/fall-2026-product-update) says customers have made more than 198,000 tool calls since the MCP server launched. More than 38,000 times, teams moved work from their AI assistants into active pursuits.

![Screenshot of the Responsive fall release news page. Under the heading "What MCP unlocks: from finding answers to acting on them", a teal box marks two paragraphs. The first gives more than 198,000 tool calls and more than 38,000 moves into active pursuits. The second lists what sellers, subject-matter experts and proposal teams can do inside their AI assistant.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/img-shot-responsive-fall-mcp.webp)

The two paragraphs on MCP actions in the fall release. We added the teal boxes. [Responsive news page, November 10, 2026, screenshot taken November 16, 2026](https://www.responsive.io/news/fall-2026-product-update)

The Responsive news release, capability page and help center each cover part of MCP. None covers all of it. Where no source gives a fact, the row says "not stated".

| Item | What the source says | Source |
| --- | --- | --- |
| Release date | November 10, 2026, Portland, Oregon | Fall release news |
| First MCP release | May 13, 2026, spring release, ChatGPT, Copilot and Claude | Spring release news |
| Supported assistants | ChatGPT, Microsoft Copilot, Claude | News, capability page |
| Actions | Draft and save responses to project questions, add comments, complete reviews and other supported actions; actions vary by assistant and user permissions | Capability page |
| Three examples in the release | A seller answers security questions and sends a follow-up; a security expert drafts and saves assigned answers; a proposal team captures, tags and routes knowledge | News |
| Permissions | The MCP server applies user-level permissions | News |
| Usage | More than 198,000 tool calls; more than 38,000 moves into active pursuits | News |
| ChatGPT setup | Create an app in ChatGPT, enter the server URL, auth type OAuth; save tool `save_project_question_response` | Help center |
| Copilot Studio setup | Add an MCP connector, enter URL and auth details, test, publish; can publish to Teams | Help center |
| Responsive as MCP client | Admin adds outside MCP connectors in Organization Settings; part of Responsive AI; contact support to enable | Help center |
| Plans and price | No public prices; which plans include MCP is not stated | Pricing page |

The other half of the fall release covers content management. The system flags duplicate and conflicting answers. It checks writing quality and metadata. It recommends whether to accept, reject or review a piece of content. Teams can automate simple decisions with rules.

The release cites a [Gartner prediction from February 2025](https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk). Through 2026, organizations will abandon 60% of AI projects that lack AI-ready data. This matters for our topic too. A clean answer library does not help if the client section of the proposal uses stale outside facts.

## Two directions: Responsive as MCP server, and Responsive as MCP client

Responsive connects to MCP in two directions. In the first direction, Responsive is the server. Claude, ChatGPT or Copilot is the client and connects to the Responsive MCP URL. The help center gives a ChatGPT example URL in the form `<tenant>.rfpio.com/oa/v1/mcp`. Each user signs in to their own Responsive account through OAuth.

In the second direction, Responsive is the client. An admin opens "Organization Settings > Responsive AI > MCP Connectors" and adds an outside MCP server. Responsive AI then calls the outside tools itself. Both directions can support account research, but the model sits in a different place.

In the first direction, the model lives in Claude or ChatGPT. It connects to both Responsive and an outside data source. It pulls client news from the data source and approved answers from Responsive. Then it saves the finished section back to the Responsive project.

In the second direction, the model lives inside Responsive. The outside data source is one connector on Responsive AI. The help center says Responsive AI uses the connector description to decide when to route a request to it. So the description should say what the connector does, for example "public company posts, open roles and news".

| Direction | Where the model runs | How to add outside data | Authentication | Write back to Responsive |
| --- | --- | --- | --- | --- |
| Responsive as server | Claude, ChatGPT or Copilot | Add a second custom MCP connector in the same assistant | OAuth for Responsive; outside source uses its own method | A tool like `save_project_question_response`, if the user can edit the project |
| Responsive as client | Responsive AI | Admin adds it under MCP Connectors with "Add New" | API key (header or query parameter) or OAuth 2.0 | Generated inside Responsive |
| No MCP | Your own script | Call the REST API | Key in a request header | Responsive API or manual paste |

The help center gives detail on API key setup for the second direction. The key can go in a request header or a query parameter. The default header name is `X-API-Key`, and you can change it. After you save, Responsive masks and encrypts the key, so you cannot see it again. An "MCP Request Preview" panel shows how Responsive will send the key.

The EveryInfra remote MCP endpoint, `https://api.everyinfra.com/mcp`, accepts only the `Authorization: Bearer <key>` header. The `?key=` form in the URL no longer works. In Responsive, you would change the header name to `Authorization` and set the value to `Bearer`, a space, and the key. We could not test whether Responsive sends that value unchanged, and the help article does not say. Check the Request Preview before you save. The OAuth path, with automatic discovery and dynamic client registration, is also untested on the Responsive side.

## What we pulled: three companies, four kinds of public data

We sent every call on November 16, 2026, between 00:18 and 00:34 Singapore time, through the [EveryInfra API](/docs#quickstart). We kept every raw response. A script recomputed every number in this article from those files. A shared test key was busy with other jobs at the time. So we hit several rate-limit responses (HTTP 429) in the first minutes and retried after the wait time in each response.

| Singapore time | Capability | Key parameters | Returned | Charged |
| --- | --- | --- | --- | --- |
| 00:23 | `linkedin/company_profile` | snowflake-computing | 1 | $0.000556 |
| 00:24 | `linkedin/company_profile` | wayfair | 1 | $0.000556 |
| 00:25 | `linkedin/jobs` | keyword Wayfair, United States, 50 | 50 | $0.000556 |
| 00:26 | `linkedin/jobs` | keyword Snowflake, United States, 50 | 50 | $0.000556 |
| 00:27 | `google_news/search` | Wayfair, last month, 20 | 20 | $0.000792 |
| 00:28 | `google_news/search` | Snowflake Inc, last month, 20 | 20 | $0.000792 |
| 00:29 | `linkedin/company_posts` | google (catalog example, for a check) | 10 | $0.002319 |
| 00:29 | `web/search` | Wayfair news November 2026 | 10 | $0.001389 |
| 00:30 | `linkedin/company_posts` | wayfair, limit 50 | 20 | $0.002319 |
| 00:30 | `linkedin/company_posts` | snowflake-computing, limit 50 | 20 | $0.002319 |
| 00:30 | `web/search` | Snowflake Inc news November 2026 | 10 | $0.001389 |
| 00:31 | `linkedin/jobs` | Snowflake company-filter URL, 50 | 50 | $0.000556 |
| 00:32 | `linkedin/company_profile` | mercadolibre | 1 | $0.000556 |
| 00:32 | `linkedin/company_posts` | mercadolibre, limit 50 | 20 | $0.002319 |
| 00:32 | `google_news/search` | Mercado Libre, Mexico, Spanish, last month | 20 | $0.000792 |
| 00:33 | `linkedin/jobs` (remote MCP) | Mercado Libre company-filter URL | 50 | $0.000556 |
| 00:34 | `linkedin/job_detail` | Wayfair role 4487569684 | 1 | $0.001194 |
| Total | 6 capabilities | 17 charged calls | 354 rows | $0.019516 |

Eight more calls failed and cost nothing. The section on failed calls covers them. The company page gives enough for the size sentence of a brief.

| Field | Wayfair | Snowflake | Mercado Libre |
| --- | --- | --- | --- |
| Employees on the page | 12,876 | 12,030 | 70,342 |
| Followers | 458,919 | 1,447,139 | 3,621,004 |
| Founded | 2002 | 2012 | not returned |
| Headquarters | Boston, MA | "The Cloud", US | Buenos Aires |
| Company IDs in `job_search_url` | 1 | 5 | 2 |

The profile also has a `job_search_url` field. The job searches below depend on it. The [LinkedIn data API](/linkedin-api) page lists all LinkedIn capabilities. See [company profile](/linkedin-api/company_profile), [company posts](/linkedin-api/company_posts), [jobs](/linkedin-api/jobs) and [job detail](/linkedin-api/job_detail) for parameters.

## Company posts: the same 20 posts can cover 9 days or 286 days

In this test, `linkedin/company_posts` returned 20 posts per company. By `posted_at`, Snowflake's 20 posts fall between November 6 and November 15, a 9-day span. All 20 are in the last 30 days. Wayfair's 20 run from September 29 to November 13, a 45-day span, with 12 in the last 30 days.

Mercado Libre's 20 posts run from January 22 to November 4, a 286-day span. Only one is in the last 30 days: the third-quarter results post on November 4. The call count and cost are the same for all three companies. The density of recent news differs by an order of magnitude.

![Stacked bar chart of 20 posts per company, split into the last 30 days (after October 17) and older. Wayfair, 20 posts over 45 days: 12 recent, 8 older. Snowflake, 20 posts over 9 days: 20 recent, 0 older. Mercado Libre, 20 posts over 286 days: 1 recent, 19 older. Recent posts are teal. Takeaway: companies that post at different rates give very different amounts of recent news for the same number of posts.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-02-posts-window-bar.en.light.svg)

After each call, we sort the posts by `posted_at`, newest first. Then we cut at October 17 and put the rest in an "older" column. A script does this step, so the model gets a list that is already sorted and cut. The model does not have to decide which post is newer. The cut removed 8 older posts for Wayfair, 0 for Snowflake and 19 for Mercado Libre. These are the "older" segments in the chart above.

We also coded post text with keyword rules. One post can count in more than one group. Wayfair's 12 recent posts include 5 on stores and logistics, 5 on people and culture, 3 on AI, 3 on events and 2 on results. The median like count is 78. The top post, with 417 likes, marks the co-founder and CEO joining a home furnishings hall of fame.

Snowflake's 20 posts include 15 on AI. The median like count is 53. The top post, with 259 likes, announces same-day availability of Claude Fable 5.2 on Cortex AI on November 10. This kind of spread tells a proposal writer what the client stresses in public this month. The proposal can open from that angle.

Mercado Libre shows a different lesson. For a company that posts rarely, company posts are poor value. Its only recent post covers the third quarter. Net revenue and financial income reached $10.75 billion, up 45% year on year, with 94 million unique active buyers. The same numbers are on its investor relations site and in its press release. For clients like this, skip posts and save $0.002319 per brief.

## Hiring signal one: a company name as keyword pulls in other employers

The `keyword` parameter of `linkedin/jobs` accepts a keyword or a full LinkedIn job search URL. We first tried the obvious form: the company name as keyword, the United States as location, and 50 results. The Wayfair search returned 50 roles, all from Wayfair. The Snowflake search returned 50 roles, and only 12 were Snowflake's own.

The other 38 came from 17 employers. Snowflake is also the name of a data platform. These employers want data engineers who know Snowflake.

| Employer in the "Snowflake" keyword results | Roles |
| --- | --- |
| First Citizens Bank | 8 |
| Tata Consultancy Services | 7 |
| Anblicks | 3 |
| Six employers with 2 each (incl. Franklin Templeton, Salesforce, Slack) | 12 |
| Eight employers with 1 each (incl. Capital One, SoFi, a university) | 8 |
| Snowflake itself | 12 |

![Stacked bar chart of four job searches with 50 roles each, split into the company's own roles and other employers. Wayfair keyword "Wayfair": 50 own, 0 other. Snowflake keyword "Snowflake": 12 own, 38 other. Snowflake company-filter URL: 50 own, 0 other. Mercado Libre filter URL via MCP: 50 own, 0 other. Own roles are teal. Takeaway: when a company name is also a skill, keyword search mixes in many other employers; the filter URL from the profile fixes it.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-03-jobs-keyword-vs-url-bar.en.light.svg)

`linkedin/company_profile` returns `job_search_url`, a job search link filtered by company IDs. Snowflake's link holds 5 company IDs, for the main page and related pages. Mercado Libre's holds 2.

We passed the Snowflake link unchanged as `keyword`. All 50 roles came back as Snowflake's own, all posted on or after October 25. The Mercado Libre call through remote MCP also returned 50 of 50.

LinkedIn's help center confirms that job search can filter by company. It also says the order considers your search activity and that search shows up to 1,000 results. So the same keyword can give a different top 50 for a different person or at a different time.

We also compared the two Snowflake searches. Of the 12 Snowflake roles from the keyword search, only 1 was in the 50 roles from the filter URL. That role was a software engineer for virtualization. Together, the two searches found 61 different Snowflake roles. So each top 50 is only a slice of the open roles. A proposal can state a share, but not a total.

The shares were close in the two samples. In the keyword set, 7 of 12 roles were engineering and AI, which is 58%. In the filter URL set, 32 of 50 were, which is 64%. So "most Snowflake hiring is for engineers" is a safe sentence. "Snowflake is hiring for 50 roles" is not.

This step costs almost nothing. A profile call costs $0.000556, and a jobs call costs the same. The saving is time spent removing other employers. It also prevents a bad sentence in the proposal, such as "you are hiring data engineers at scale", when the roles belong to someone else.

## Hiring signal two: what the role mix says, and where it misleads

With clean lists, we put each role into one group by keywords in its title. The groups were sales and presales, engineering and AI, stores and operations, product, marketing, analytics, finance and HR and legal, and other. We wrote the rules after we looked at the raw Wayfair and Snowflake lists, and we did not change them after that.

The rules run in order, and the first match wins. A title with words such as sales, account executive, solution engineer, venta or comercial goes to sales. A title with store, retail, warehouse or operations goes to stores. A title with engineer, scientist, software or AI goes to engineering. A title with no match goes to other.

![Grouped bar chart of 50 roles per company by function. Sales, presales, GTM: Wayfair 6, Snowflake 8, Mercado Libre 26. Engineering, data engineering, AI: 2, 32, 10. Stores, warehouses, operations: 29, 0, 5. Product: 1, 7, 1. Marketing: 5, 2, 2. Analytics: 3, 0, 2. Finance, HR, legal: 3, 1, 4. Other: 1, 0, 0. Mercado Libre is teal. Takeaway: Wayfair hires mostly for stores and warehouses, Snowflake for engineering and AI, and Mercado Libre for field and account sales; three of Wayfair's six sales roles are store roles and count as misreads.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-04-job-functions-bar.en.light.svg)

The three companies look very different. Wayfair has 29 of 50 roles in stores, warehouses and operations. That matches its posts about new stores. On October 7 it announced its 11th store, in Cherry Hill, New Jersey. On November 12 it announced a second New York store in the Buffalo area for 2028, at 96,000 square feet.

Snowflake has 32 of 50 roles in engineering, data engineering and research, plus 7 product roles. Its 8 sales roles include Spanish-speaking and French-speaking sales development representatives and an enterprise account executive for Norway. You can read that as growth in Europe and Spanish-speaking markets. Mercado Libre has 26 of 50 roles in sales. Most are Mercado Pago field sales, payment terminal and direct sales roles in Chile and Mexico.

Here are Wayfair's six "sales" roles. Three are "Assistant Store Manager of Sales", "Assistant Manager, Retail Sales" and "Assistant Store Manager, Retail Sales". These are store management roles. The sales rule runs before the stores rule, so "Retail Sales" matched sales first. Only three relate to business customers. They are a commercial go-to-market role for Wayfair Professional, a B2B sales strategy analytics manager and a sales compensation analyst.

A team that sells to Wayfair's business buying side could read "six sales roles" as fast B2B growth. Read each title, and the claim shrinks by half.

A job detail call can confirm one signal. We called `linkedin/job_detail` for the Wayfair commercial role, ID 4487569684, posted November 4. It cost $0.001194. The description says Wayfair Professional is the company's B2B business. It is "a multi-billion dollar business serving 1M+ active customers" in office, interior design, contractor and property management. The role owns go-to-market strategy for property management and works from the Boston headquarters four days a week.

A team that sells office furniture procurement software or property sector data can quote this in a proposal.

Read the job data with care in four more places. Wayfair's 50 rows hold only 45 distinct titles, and Mercado Libre's 50 hold only 38. The same role appears once per city. Remove duplicates by title, or by title and location, before you count roles. The choice depends on whether you want to talk about functions or headcount.

Wayfair's oldest role dates from May 27, and only 23 of its 50 rows are from the last 30 days. Long-running roles dilute "recent hiring".

The `applicant_count` column also needs care. Wayfair has 26 rows at exactly 25 and 2 at 200. Snowflake has 21 rows at 200. We think these are display buckets such as "first 25 applicants" or "over 200 applicants", so they are not exact counts. Salary is common in US rows: 45 of Wayfair's 50 have it. Only 1 of Mercado Libre's 50 Latin American rows has a salary.

## News and web search: check every date

`google_news/search` with a one-month window returned 20 rows per company. All 60 rows across the three companies had a date, and all fell within 30 days. Wayfair's 20 included 5 from Yahoo Finance plus CNBC, CBS and several Buffalo local stations. The topics were third-quarter results, the Buffalo store and the late-October Way Day sale.

Snowflake's 20 included 11 from Yahoo Finance and 7 from Snowflake's own press releases and blog. For Mercado Libre we searched in Spanish for Mexico. Two of the 20 rows used "mercado libre" in the sense of "free market" and had nothing to do with the company. A person had to remove them. See the [Google News API](/google-news-api) and [news search](/google-news-api/search) pages for parameters.

| News set | Rows | Within 30 days | Largest source | Company's own pages | Off-topic |
| --- | --- | --- | --- | --- | --- |
| Wayfair, US, English | 20 | 20 | Yahoo Finance, 5 | 0 | 0 |
| Snowflake Inc, US, English | 20 | 20 | Yahoo Finance, 11 | 7 | 0 |
| Mercado Libre, Mexico, Spanish | 20 | 20 | Milenio, 3 | 0 | 2 |

`web/search` results were messier. We searched "<company> news November 2026" for 10 rows each. Wayfair's 10 included 3 with no date, 1 with only "1 year ago", and a layoffs story from August 22, 2022. Snowflake's 10 included 1 with no date and 2 with only "days ago". The field note already warns that `published_at` for web search can be missing or wrong.

That matches what Google says about its own results. Google estimates the date in search results from several signals on the page. If the page does not mark its date clearly, the estimate can be wrong. See the [web data API](/web-api) and [web search](/web-api/search) pages for parameters.

![Stacked bar chart of four result sets by date quality. Wayfair google_news, 20 rows: 20 in the last 30 days. Snowflake google_news, 20 rows: 20 in the last 30 days. Wayfair web/search, 10 rows: 2 recent, 4 older, 1 relative date, 3 no date. Snowflake web/search, 10 rows: 6 recent, 1 older, 2 relative dates, 1 no date. Recent rows are teal. Takeaway: news results have clean dates; web results need a manual date check.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-06-date-quality-bar.en.light.svg)

So we split the work between the two sources. News search answers "what happened this month", and its dates can be used directly. Web search finds primary pages, such as investor relations pages and original press releases. We open each page, check the date and the numbers, and only then quote it.

If a model takes the August 2022 layoffs story as recent news, a November 2026 proposal will say "you are cutting staff". The story is more than four years old.

## Writing the section: two sample "our understanding" drafts

We built two sample drafts from the data above. In the first, we write to Wayfair's corporate purchasing team. In the second, we write to Snowflake's sales operations team. Each sentence ends with a source ID in brackets: a post ID, a role ID, or a source and date for news. A model wrote the drafts under fixed rules. One rule mattered most: a sentence without a source ID does not go into the draft.

Wayfair draft. Your third-quarter 2026 revenue was $3.4 billion, up 9.7%, with US revenue up 11.4% and 22.2 million active customers (investor relations, November 4). Your store rollout is speeding up. In October you announced your 11th store in Cherry Hill (post 7513659797682982912, October 7, older than 30 days). On November 12 you announced a second New York store in the Buffalo area for 2028 (post 7526750679131062272). Of your 50 open roles, 29 are in stores, warehouses and operations.

Wayfair Professional, your business customer unit, serves more than 1 million active business customers and is hiring a commercial lead for property management (role 4487569684, November 4). You use AI at scale inside the company. More than 1,600 engineers use an AI coding tool (post 7526313811780472832, November 11). AI has validated more than 11 million product specs and resolves more than 41,000 supplier tickets each month (post 7524124393464217600, November 5).

Snowflake draft. At Expedition on November 3, you launched Business Ontology and Cortex AI Gateway, so AI agents can work inside governed data and permissions (press release, November 3; post 7524576411794194432, November 6). On November 10 you made Claude Fable 5.2 available the same day in private preview on Cortex AI (post 7525739668739497984). A post on November 14 proposed "cost per correct, governed outcome" as the way to measure AI spend (post 7527181045956349952).

In hiring, 32 of your 50 recent open roles are in engineering, data engineering and research. Your sales roles include Spanish-speaking and French-speaking representatives and a Norway account executive (company-filter URL, pulled November 16 at 00:31 Singapore time). Your third-quarter results come out on December 2 (press release, November 2). Any budget-cycle point in the proposal should take that date into account.

| Fact in the draft | Source type | Source ID | Date | Within 30 days |
| --- | --- | --- | --- | --- |
| Wayfair Q3 revenue $3.4 billion, up 9.7% | Investor relations | investor.wayfair.com Q3 results | 2026-11-04 | Yes |
| Wayfair 11th store in Cherry Hill | Company post | 7513659797682982912 | 2026-10-07 | No, flagged |
| Wayfair Buffalo store in 2028 | Company post | 7526750679131062272 | 2026-11-12 | Yes |
| Wayfair Professional, 1M+ active business customers | Job detail | 4487569684 | 2026-11-04 | Yes |
| 1,600+ Wayfair engineers use an AI coding tool | Company post | 7526313811780472832 | 2026-11-11 | Yes |
| Snowflake Business Ontology and AI Gateway | Press release | snowflake.com press release | 2026-11-03 | Yes |
| Snowflake adds Claude Fable 5.2 | Company post | 7525739668739497984 | 2026-11-10 | Yes |
| Snowflake Q3 results on December 2 | Press release | snowflake.com press release | 2026-11-02 | Yes |

Neither draft quotes a web search row. In this test, web search only helped us find the investor relations page and the press releases. We opened those pages, checked them, and quoted the pages themselves.

## Keep client-supplied material apart from public data

A proposal writer usually holds three kinds of material at once. There is the RFP and its attachments, there are notes from discovery calls, and there is public data like the data above. The first two come from the client and often carry confidentiality terms. The third is open to anyone.

If you mix all three in one prompt and ask for "recent context", two problems are common. The model can present a private remark from a call as "public reports say". It can also blend a public number with a number the client gave verbally, so neither source matches.

| Aspect | Public data | Client-supplied material |
| --- | --- | --- |
| Examples | Company posts, open roles, press releases, earnings | RFP, attachments, call notes, client emails |
| Storage | A separate folder in the proposal project, filed by pull time | Responsive project attachments, under project permissions |
| How it reaches the model | Only structured results with source IDs | A separate pass, marked "client-supplied, do not quote externally" |
| How the proposal cites it | "In your November 12 announcement" | "As stated in your requirements document" |
| Can it appear in material for third parties | Yes, with the source | No, unless the contract allows it |
| Expiry | Flag after 30 days; drop from the section after 90 days | The client's latest version wins |

We follow three rules. First, the pass that pulls public data gets no client material in its prompt. A reviewer then reads the draft next to the client material and can see where each sentence came from. Second, every sentence carries a source ID. Client sentences use "file name and page", and public sentences use a post, role or URL. The two formats look different, so a mix-up is easy to see.

Third, never put client material into the parameters of an outside search or data API. For example, do not search the news for an unreleased project name that a client mentioned on a call. That sends confidential information to a third party.

Personal data needs its own line. Company posts and job pages sometimes name employees, for example a warehouse manager in a feature post. A proposal does not need any of those names. When you must refer to a person, use the job title, for example "your CFO said in a November interview with Bloomberg".

On November 10, 2026, the European Data Protection Board adopted [version 2.0](https://www.edpb.europa.eu/system/files/2026-11/edpb_guidelines_202401_legitimateinterest_v2_en.pdf) of its guidelines on legitimate interest as a legal basis. A controller must assess and document three conditions before processing. The interest must be legitimate. The processing must be necessary. The person's rights must not override the interest.

Account research that collects only company-level facts avoids most of this work. This is a work habit, not legal advice. Ask your own counsel about specific cases.

## Where hiring signals mislead

| Case | Example in this test | Effect | How to avoid it |
| --- | --- | --- | --- |
| Company name is also a skill | Keyword "Snowflake": 38 of 50 roles from other employers | Other employers' hiring is credited to the client | Use `job_search_url` from the profile |
| Store roles with "Sales" in the title | 3 of Wayfair's 6 "sales" roles are store managers | B2B sales growth looks bigger than it is | Read each title after the rule runs |
| Same role in many cities | Wayfair: 45 titles in 50 rows; Mercado Libre: 38 in 50 | Headcount looks bigger | Remove duplicates by title, or title and location |
| Roles that stay open for months | Wayfair's oldest is May 27; only 23 rows are recent | A steady state looks like a new move | Use the last 30 days, or compare with last month |
| Applicant counts are display buckets | Wayfair: 26 rows at 25; Snowflake: 21 rows at 200 | A bucket looks like demand | Treat as a rough signal; keep it out of the proposal |
| Affiliates counted together | Snowflake's filter URL holds 5 company IDs | Subsidiaries and acquired teams blend into the parent | Split by `company_name` if needed |
| Hiring is not buying | Mercado Libre's 26 sales roles are mostly field sales | It looks like a buyer of sales tools | Hiring shows direction; budget needs other evidence |

Mercado Libre has 26 sales roles in its 50. By business line, Mercado Pago has 15, Fintech has 7, and Mercado Ads and Marketplace have 2 each. By country, Chile has 14 of the 26 and Mexico has 9. Mercado Libre is hiring Mercado Pago field sellers in large numbers. You can read that as a push for offline merchants in Chile and Mexico. It does not mean the company will buy a CRM or sales training.

Keep "what the client is doing" apart from "what the client needs from us". The second claim needs the client's own words, from the RFP or a call. That brings us back to the previous section: those words are client material and go in the other column.

The [US Bureau of Labor Statistics](https://www.bls.gov/news.release/jolts.nr0.htm) reported on November 3 that job openings were about 7.5 million in September. On November 12, Indeed Hiring Lab published quarterly reports on [tech](https://hiringlab.indeed.com/2026/11/12/2026-q3-us-tech-labor-market-update/) and [retail](https://hiringlab.indeed.com/2026/11/12/2026-q3-us-retail-labor-market-update/) hiring. Retail seasonal postings grew faster than in the same period of earlier years. Before you compare one company's role count with figures like these, check the scope. National data estimates all postings across channels. Our data is the top 50 open roles of one company on one platform at one moment.

## One remote MCP round trip: tools/list and tools/call

Inside an assistant, all of these calls run through one tool. At 00:33 Singapore time we sent a `tools/list` request to `https://api.everyinfra.com/mcp`. The tool list included `everyinfra_list_capabilities` for the catalog, `everyinfra_call_api` for any capability, and `everyinfra_get_job` for async jobs. The tool description says the catalog covers 106 platforms and 574 capabilities. Then we used `everyinfra_call_api` to pull Mercado Libre's open roles, with the filter URL from its profile as the parameter.

```bash
# The tools/call sent at 00:33:36 Singapore time (the key comes from an environment variable only)
curl -s https://api.everyinfra.com/mcp \
  -H "Authorization: Bearer $EVERYINFRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"everyinfra_call_api",
       "arguments":{"platform":"linkedin","action":"jobs",
       "params":{"keyword":"https://www.linkedin.com/jobs/search?geoId=92000000&f_C=14226%2C107272452","limit":50}}}}'
```

![Screenshot of one remote MCP call. The top half is the JSON-RPC request to https://api.everyinfra.com/mcp: method tools/call, tool everyinfra_call_api, linkedin / jobs, a job search URL with company IDs as keyword, limit 50. The bottom half is an excerpt of the response with count 50. A teal box marks two results: a Mercado Pago field sales role in Monterrey posted October 19 with 74 applicants, and a Mercado Pago field sales role in Temuco posted October 29 with 75 applicants. At the bottom, billing shows charged true and charged_usd 0.000556.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/img-shot-mcp-jobs-response.webp)

A remote MCP call sent at 00:33 Singapore time on November 16, 2026. The response holds 50 roles and the charge for the call. [EveryInfra remote MCP response, November 16, 2026, 00:33 SGT, screenshot](https://api.everyinfra.com/mcp)

The `content` field of the response holds a JSON text. Its `results` hold 50 roles, and `billing` shows a charge of $0.000556, the same as a direct REST call. In the MCP specification, `tools/list` lets a client discover the tools on a server, and `tools/call` runs one. JSON Schema describes tool inputs and outputs.

Remote servers use Streamable HTTP, with one endpoint for POST and GET. The authorization section of the spec recommends OAuth 2.1 for HTTP transports and also allows other methods. Our call used a bearer token, as defined for OAuth 2.0 in RFC 6750. Setup details are on the [MCP server page](/mcp-server).

## Setup in Claude, ChatGPT, Copilot and Responsive (configuration examples)

The four setups below are configuration examples. Each one follows the vendor's help center. We tested only the `tools/list` and `tools/call` requests that we sent with curl in the previous section. We did not test the setup in Claude, ChatGPT, Copilot or Responsive.

A Claude custom connector connects from Anthropic's cloud to your MCP server, so the server must be reachable from the public internet. On Team and Enterprise plans, an owner adds the connector and each member connects on their own. Connectors inherit each user's permissions in the connected system.

```text
{
  "mcpServers": {
    "everyinfra": {
      "url": "https://api.everyinfra.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    },
    "responsive": {
      "url": "https://<tenant>.rfpio.com/oa/v1/mcp"
    }
  }
}
```

This configuration connects one assistant to two servers. EveryInfra supplies public data. Responsive supplies internal answers and takes the write-back. The Responsive entry uses OAuth, as its help center describes, so the config holds no key. The first use opens a sign-in page.

In ChatGPT, the OpenAI help center explains how to add a custom MCP server in settings. The Responsive example for ChatGPT uses OAuth, and its save tool is `save_project_question_response`. In Copilot Studio, the Microsoft docs say to add a new tool and choose Model Context Protocol. You enter a server name, a description and a URL. Authentication can be none, API key or OAuth 2.0. Copilot Studio supports only the Streamable transport and dropped SSE after August 2025.

To add EveryInfra to Responsive as an outside connector, the help center steps are as follows. Open Organization Settings, then Responsive AI, then MCP Connectors, then Add New. Enter a name and a description, such as "Look up public company pages, company posts, open roles and news, with dates and source links". Enter `https://api.everyinfra.com/mcp` as the URL. Choose API key, choose Request Header, and change `X-API-Key` to `Authorization`. Enter `Bearer` and the key as the value, and check the Request Preview before you save.

Responsive then discovers the tools and turns all of them on. An admin can keep only `everyinfra_call_api` and `everyinfra_list_capabilities`. The help center also gives the lookup order. Responsive AI checks the content library first, then outside sources, and the MCP server last, unless the prompt names a tool. So the prompt should say "use everyinfra_call_api to query linkedin/jobs".

Whatever the setup, write the rules into the prompt. Sort by date and keep the last 30 days. Give each sentence a source ID. Remove duplicate roles by title. Use web search only to find original pages. The snippet below holds our sort and cut logic. A model can follow it as an instruction, and a script can run it.

```text
from datetime import datetime, timezone, timedelta
cut = datetime(2026, 11, 16, tzinfo=timezone.utc) - timedelta(days=30)
posts = sorted(resp["results"], key=lambda p: p["posted_at"], reverse=True)   # sort newest first
recent = [p for p in posts if datetime.fromisoformat(p["posted_at"]) >= cut]
older = [p for p in posts if datetime.fromisoformat(p["posted_at"]) < cut]    # separate column, not in the brief
seen, jobs = set(), []
for j in job_resp["results"]:
    key = (j["title"].strip().lower(), j.get("location"))                     # one row per title and city
    if key not in seen:
        seen.add(key); jobs.append(j)
```

## What a sales proposal AI agent pays for research per proposal

One client brief, with one call per capability, costs the following. Company page $0.000556, posts $0.002319, open roles $0.000556, job detail $0.001194, news $0.000792 and web search $0.001389. The total is $0.006806. If you quote three job details, add two more detail calls for a total of $0.009194.

At 100 proposals a year, the data cost is about $0.68. Model costs come on top and depend on the assistant you use. Prices follow the [pricing page](/pricing) and the `price_usd` of each capability in the catalog. In this test, the `charged_usd` of all 17 calls matched the catalog prices.

![Horizontal bar chart of the data cost of one client brief by capability, in US dollars. linkedin/company_posts 0.002319, web/search 0.001389, linkedin/job_detail 0.001194, google_news/search 0.000792, linkedin/company_profile 0.000556, linkedin/jobs 0.000556; total 0.006806. Company posts are teal. Takeaway: one call per capability costs less than one cent, company posts are about a third of it, and 100 proposals cost about 0.68 US dollars.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-05-brief-cost-bar.en.light.svg)

| Option | Calls per client | Cost per client | 100 proposals |
| --- | --- | --- | --- |
| Lean: profile, roles, news | 3 | $0.001904 | about $0.19 |
| Standard: one call per capability | 6 | $0.006806 | about $0.68 |
| With job details: quote 3 roles | 8 | $0.009194 | about $0.92 |
| Client that rarely posts: no posts | 5 | $0.004487 | about $0.45 |

We measured time from the call log. From the first request at 00:18:57 to the last at 00:34:21, the test took 15 minutes and 24 seconds. For the first 4 minutes, we waited on rate limits for the shared key. Between 00:24 and 00:29, eight async calls failed and we retried them. After we changed to sync mode, 11 charged calls ran from 00:29:42 to 00:34:21. That is 4 minutes and 39 seconds, or about 25 seconds per call.

We did not time the model draft or the human check, so we give no number for them. In the check, open the source for every number that goes into the proposal: revenue, store count and role count.

## The 8 failed calls: async mode returned 503 and cost nothing

Eight calls failed in this test. We sent `linkedin/company_posts` and `web/search` in async mode (`"mode":"async"`) twice each for Wayfair and Snowflake. All eight returned HTTP 503. The posts calls said the data was unavailable for now. The web search calls reported an internal error while preparing the request.

The posts responses had a `billing` object with `charged` set to false and an amount of 0. They also had a refund status. The web search responses had no `billing` object, and `next_steps` said "Not charged". No call in either group was charged. We switched to synchronous mode, and all four calls succeeded with 20 or 10 rows each.

| Capability | Mode | Attempts | Result | Charged |
| --- | --- | --- | --- | --- |
| `linkedin/company_posts` | async | 2 each for Wayfair and Snowflake | 4 × 503 | $0 |
| `web/search` | async | 2 each for Wayfair and Snowflake | 4 × 503 | $0 |
| `linkedin/company_posts` | sync | 1 each for Wayfair, Snowflake, Mercado Libre | 20 rows | $0.002319 each |
| `web/search` | sync | 1 each for Wayfair and Snowflake | 10 rows | $0.001389 each |

To diagnose, we first called company posts in sync mode with the catalog's example parameter, the company "google". It returned 10 posts for $0.002319. That confirmed the capability worked, and then we went back to our companies. The check is included in the 17 charged calls.

If an assistant hits a 503 halfway, the model may say "no recent activity found". That sentence in a proposal would be false.

When the API cannot get results, it returns 503 and charges nothing. The response has `retryable` and `retry_after`. For the posts calls, `retry_after` was 60 seconds with up to 2 retries. For the web search calls, it was 30 seconds with 1 retry. Tell the model in the prompt that 503 means "retry after `retry_after`, or use sync mode". It must never mean "the company has no recent news".

## Industry moves: proposal tools added MCP this month

Two weeks before the Responsive fall release, [QorusDocs launched QPilot Workspace](https://www.usatoday.com/press-release/story/52675/qorusdocs-launches-qpilot-workspace-where-people-and-ai-agents-build-winning-proposals-together/) on October 27, 2026. Its press release says MCP connectors link it to Microsoft Copilot, Claude and ChatGPT. It also cites the company's survey: 68% of respondents saw proposal volume rise.

AdExchanger reported in November that TIME's ad sales team uses an AI agent to read RFPs and draft proposals. The story mentions an MCP server. AutoRFP published a November post on using Claude connectors for RFPs.

![Timeline scatter chart from September 2024 to January 2027. 2024-11-25 Anthropic releases MCP. 2025-11-25 MCP spec version 2025-11-25. 2026-05-13 Responsive spring release adds MCP. 2026-10-27 QorusDocs adds MCP connectors. 2026-11-10 Responsive fall release expands MCP. Responsive's two dots are teal. Takeaway: Responsive's two releases are 181 days apart, and a peer added MCP in between.](https://everyinfra.com/articles/responsive-mcp-sales-proposal-account-research/fig-07-mcp-timeline.en.light.svg)

Over a longer span, Anthropic released MCP on November 25, 2024. The latest spec version is dated 2025-11-25. Responsive's spring release (2026-05-13) and fall release (2026-11-10) are 181 days apart. The [spring release](https://www.responsive.io/news/spring-2026-product-update) cites a Gartner prediction. By 2027, 80% of AI agent interaction with SaaS apps will run through SaaS-hosted MCP servers, up from less than 5% in 2025.

The [2025 State of Strategic Response Management report](https://www.responsive.io/resources/whitepapers/srm-report-2025) from Responsive and APMP has one useful number. 75% of organizations say buyers demand faster responses and more personalization under tighter budgets.

[APMP](https://www.apmp.org/Web/Web/About-Us/Winning-Business-Ecosystem.aspx) splits winning business into a pre-bid capture stage and a bid and proposal stage. Analysis of client objectives, challenges and history belongs to the capture stage. Writing starts after the RFP arrives. So do account research during capture, before the RFP. When you write, refresh only the last week.

## Limits: what this method cannot do

First, public data shows what a client does, not what it needs. Hiring, posts and news are things the client chose to say in public, with a PR filter. The "our understanding" section can use them. The "your challenges" section must rest on what the client said in the RFP and in calls.

Second, each call returned 20 posts and 50 roles. For companies that post and hire a lot, that is only a slice. "You are hiring for 50 roles" is wrong. "Among roles posted in the last 30 days" is right.

Third, LinkedIn's official API does not cover this kind of reading. Microsoft's documentation describes the Community Management API for pages where you hold an admin role. Reading posts, posting and analytics apply only to organizations you manage. In proposal work, you need the client's page, and you are not its admin.

Fourth, a person must review every draft. Our drafts follow fixed rules, and the rules miss some cases, such as odd job titles or a post that covers two topics. Only a person catches those. Fifth, this article is not legal advice. Ask your own counsel about cross-border use of public data and about quoting third-party reporting in proposals.

## Get started

Step one: get an API key from the [EveryInfra quickstart](/docs#quickstart). Call `linkedin/company_profile` for a client you know well and note the `job_search_url` in the response.

Step two: pass that URL to `linkedin/jobs`. Then call `linkedin/company_posts` and `google_news/search`. Sort the results by date, keep the last 30 days, and read them yourself. Confirm that no other employer's roles slipped in.

Step three: connect the [remote MCP server](/mcp-server) in your assistant. Put the rules from step two in the prompt. Sort, cut, remove duplicates, cite a source ID and retry on 503.

Step four: with Responsive, use the "Responsive as server" direction and connect both servers in one assistant. As an alternative, an admin can add an outside connector under MCP Connectors in Responsive. Check the Request Preview before you save.

Step five: before the proposal goes out, a person checks three kinds of numbers in the client section. These are revenue and results, store or employee counts, and role counts.

For more ways to read hiring data, see [job posting data as buying signals](/blog/job-posting-data-buying-signals). For another case of public data in a business tool through remote MCP, see [ActiveCampaign MCP: subject lines and segments from real reviews](/blog/activecampaign-mcp-active-intelligence-social-review-data).

## FAQ

### What is the Responsive MCP server, and what can it do in an assistant?

The Responsive MCP Server connects ChatGPT, Microsoft Copilot and Claude to approved answers and project data in Responsive. Its capability page says users can draft and save responses to project questions, add comments and complete reviews. Available actions vary by assistant and user permissions. The fall release on November 10, 2026, extended the server from lookups to these actions.

### Can Responsive (responsive io MCP) connect to an outside data source?

Yes. A help article covers connecting outside MCP servers to Responsive AI. An admin adds the server under MCP Connectors in Organization Settings.

Authentication supports API keys, in a header or a query parameter, and OAuth 2.0. Responsive discovers the tools on its own. The MCP client feature is part of Responsive AI, and the help article says to contact Responsive support to enable it. It does not say which plans include it.

### What does data cost for a sales proposal AI agent?

In our test, one client brief with one call per capability costs $0.006806. Profile, roles and news alone cost $0.001904. For 100 proposals a year, the range is $0.19 to $0.92, depending on which calls you make. Model costs come on top and depend on your assistant and plan.

### Can I use ChatGPT as a sales proposal AI agent?

Yes. ChatGPT supports custom MCP servers, and Responsive documents a ChatGPT setup. The method is the same as in Claude: connect two servers in one assistant, one for public data and one for Responsive. Put the rules for sorting, duplicates and source IDs in the prompt. Assistants differ a little in how they call tools. On first use, ask the assistant to list the parameters of each tool call, and check them.

### How does this fit with client records in Salesforce?

Opportunity records, contacts and call notes in Salesforce are client material. They go in the client-supplied column and stay out of the public-data prompt. After you pull public data, you can write fields such as company size, headquarters and recent role count back to the CRM account record. Add the pull date, and pull again after 30 days.

### How often should I refresh hiring data?

Pull once when you write the proposal. If news breaks in the last week before you submit, refresh the news only. In this test, all 50 Snowflake roles were posted on or after October 25, so roles change fast. Wayfair has long-running store roles, with the oldest from May 27. For a quarterly account review, pull roles once a month and compare the mix by function. The change tells you more than a single count.

## References

1. [Responsive, Fall 2026 Release Expands MCP Capabilities, November 10, 2026](https://www.responsive.io/news/fall-2026-product-update), read and screenshot November 16, 2026
1. [Responsive, Spring 2026 Release, May 13, 2026](https://www.responsive.io/news/spring-2026-product-update), read November 16, 2026
1. [Responsive Blog, How to use Responsive in ChatGPT, Copilot and Claude](https://www.responsive.io/blog/how-to-use-responsive-chatgpt-copilot-claude), read November 16, 2026
1. [Responsive, MCP Server capability page](https://www.responsive.io/capability/responsive-mcp-server), read November 16, 2026
1. [Responsive Blog, Responsive MCP Server: new LLM integrations](https://www.responsive.io/blog/responsive-mcp-server-new-llm-integrations), read November 16, 2026
1. [Responsive Help Center, Setting up MCP connectors in Responsive](https://help.responsive.io/en-US/responsive/article/PKB9b47g-setting-up-mcp-connectors-in-responsive), read November 16, 2026
1. [Responsive Help Center, Responsive MCP Server FAQs](https://help.responsive.io/en-US/responsive/article/ElYb6CUB-responsive-mcp-server-faqs), read November 16, 2026
1. [Responsive Help Center, Setting up Responsive MCP server in Copilot Studio](https://help.responsive.io/en-US/responsive/article/monPeZaE-setting-up-responsive-mcp-server-in-copilot-studio), read November 16, 2026
1. [Responsive Help Center, Using ChatGPT with Responsive AI](https://help.responsive.io/en-US/responsive/article/2FyiH7XT-using-chatgpt-with-responsive-ai), read November 16, 2026
1. [Responsive Help Center, Connecting a Seismic MCP server to Responsive](https://help.responsive.io/en-US/responsive/article/EHOmzcXT-connecting-a-seismic-mcp-server-to-responsive), read November 16, 2026
1. [Responsive Help Center, Connecting Responsive MCP server to generative AI tools](https://help.responsive.io/en-US/responsive/article/duV3ckq5-connecting-responsive-mcp-server-to-generative-ai-tools), read November 16, 2026
1. [Responsive, Pricing](https://www.responsive.io/pricing), read November 16, 2026
1. [Responsive, Responsive AI product page](https://www.responsive.io/product/ai), read November 16, 2026
1. [Responsive and APMP, 2025 State of Strategic Response Management Report](https://www.responsive.io/resources/whitepapers/srm-report-2025), read November 16, 2026
1. [Gartner, Lack of AI-Ready Data Puts AI Projects at Risk, February 26, 2025](https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk), read November 16, 2026
1. [Anthropic, Introducing the Model Context Protocol, November 25, 2024](https://www.anthropic.com/news/model-context-protocol), read November 16, 2026
1. [MCP specification 2025-11-25: Tools](https://modelcontextprotocol.io/specification/2025-11-25/server/tools), read November 16, 2026
1. [MCP specification 2025-11-25: Authorization](https://modelcontextprotocol.io/specification/2025-11-25/basic/authorization), read November 16, 2026
1. [MCP specification 2025-11-25: Transports](https://modelcontextprotocol.io/specification/2025-11-25/basic/transports), read November 16, 2026
1. [GitHub, modelcontextprotocol specification repository](https://github.com/modelcontextprotocol/modelcontextprotocol), read November 16, 2026
1. [Claude Help Center, Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp), read November 16, 2026
1. [Claude Help Center, Use connectors to extend Claude's capabilities](https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities), read November 16, 2026
1. [OpenAI Help Center, Connecting custom MCP servers in ChatGPT](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt), read November 16, 2026
1. [OpenAI Developers, Build an MCP server](https://developers.openai.com/apps-sdk/build/mcp-server), read November 16, 2026
1. [Microsoft Learn, Extend your agent with Model Context Protocol](https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-extend-action-mcp), read November 16, 2026
1. [Microsoft Learn, Connect your agent to an existing MCP server](https://learn.microsoft.com/en-us/microsoft-copilot-studio/mcp-add-existing-server-to-agent), read November 16, 2026
1. [IETF, RFC 6750: OAuth 2.0 Bearer Token Usage](https://datatracker.ietf.org/doc/html/rfc6750), read November 16, 2026
1. [EveryInfra capability catalog](https://api.everyinfra.com/api/v1/social/catalog), read November 16, 2026, 00:15
1. [Microsoft Learn, LinkedIn Community Management API overview](https://learn.microsoft.com/en-us/linkedin/marketing/community-management/community-management-overview?view=li-lms-2026-10), read November 16, 2026
1. [Microsoft Learn, LinkedIn Posts API](https://learn.microsoft.com/en-us/linkedin/marketing/community-management/shares/posts-api?view=li-lms-2026-10), read November 16, 2026
1. [Microsoft Learn, LinkedIn Organization Access Control by Role](https://learn.microsoft.com/en-us/linkedin/marketing/community-management/organizations/organization-access-control-by-role?view=li-lms-2026-10), read November 16, 2026
1. [LinkedIn Help, Filter and sort job search results](https://www.linkedin.com/help/linkedin/answer/a507441/filter-and-sort-job-search-results), read November 16, 2026
1. [LinkedIn Help, Sort your feed by top or recent](https://www.linkedin.com/help/linkedin/answer/a1480504), read November 16, 2026
1. [Google Search Central, Influencing publication dates in search results](https://developers.google.com/search/docs/appearance/publication-dates), read November 16, 2026
1. [Wayfair Investor Relations, Third Quarter 2026 Results, November 4, 2026](https://investor.wayfair.com/news/news-details/2026/Wayfair-Announces-Third-Quarter-2026-Results-Reports-Double-Digit-U-S--Revenue-Growth/default.aspx), read November 16, 2026
1. [Wayfair Investor Relations, Wayfair Schedules Third Quarter 2026 Earnings Release](https://investor.wayfair.com/news/news-details/2026/Wayfair-Schedules-Third-Quarter-2026-Earnings-Release-and-Conference-Call/default.aspx), read November 16, 2026
1. [About Wayfair, Wayfair to open a store in Buffalo, New York](https://www.aboutwayfair.com/category/company-news/wayfair-to-open-a-store-in-buffalo-new-york), read November 16, 2026
1. [Snowflake press release, Expedition 2026: Business Ontology and AI Gateway, November 3, 2026](https://www.snowflake.com/en/news/press-releases/expedition-2026-business-ontology-ai-gateway/), read November 16, 2026
1. [Snowflake press release, Q3 fiscal 2027 results date, November 2, 2026](https://www.snowflake.com/en/news/press-releases/fy27-q3-earnings-date-announcement/), read November 16, 2026
1. [Business Wire, MercadoLibre Reports Third Quarter 2026 Financial Results, November 4, 2026](https://www.businesswire.com/news/home/20261104157063/en/MercadoLibre-Inc.-Reports-Third-Quarter-2026-Financial-Results), read November 16, 2026
1. [Mercado Libre Investor Relations, News and Events](https://investor.mercadolibre.com/news-and-events), read November 16, 2026
1. [Office of the President of Mexico, Mercado Libre robotic distribution center opens, November 3, 2026](https://www.gob.mx/presidencia/prensa/presidenta-claudia-sheinbaum-inaugura-el-primer-centro-de-distribucion-robotico-de-mercado-libre-en-hidalgo-como-parte-del-plan-mexico?idiom=es), read November 16, 2026
1. [US Bureau of Labor Statistics, Job Openings and Labor Turnover Summary, November 3, 2026](https://www.bls.gov/news.release/jolts.nr0.htm), read November 16, 2026
1. [Indeed Hiring Lab, 2026 Q3 US Tech Labor Market Update, November 12, 2026](https://hiringlab.indeed.com/2026/11/12/2026-q3-us-tech-labor-market-update/), read November 16, 2026
1. [Indeed Hiring Lab, 2026 Q3 US Retail Labor Market Update, November 12, 2026](https://hiringlab.indeed.com/2026/11/12/2026-q3-us-retail-labor-market-update/), read November 16, 2026
1. [Acquisition.gov, FAR 15.305 Proposal evaluation](https://www.acquisition.gov/far/15.305), read November 16, 2026
1. [APMP, Winning Business Ecosystem](https://www.apmp.org/Web/Web/About-Us/Winning-Business-Ecosystem.aspx), read November 16, 2026
1. [EUR-Lex, Regulation (EU) 2016/679 (GDPR)](https://eur-lex.europa.eu/eli/reg/2016/679/oj), read November 16, 2026
1. [EDPB, Guidelines 1/2024 on Article 6(1)(f) GDPR, Version 2.0, November 10, 2026](https://www.edpb.europa.eu/system/files/2026-11/edpb_guidelines_202401_legitimateinterest_v2_en.pdf), read November 16, 2026
1. [QorusDocs press release, QPilot Workspace, October 27, 2026](https://www.usatoday.com/press-release/story/52675/qorusdocs-launches-qpilot-workspace-where-people-and-ai-agents-build-winning-proposals-together/), read November 16, 2026
1. [AdExchanger, TIME's AI agents do the coordinating and its sellers still do the selling](https://www.adexchanger.com/ai/times-ai-agents-do-the-coordinating-and-its-sellers-still-do-the-selling/), read November 16, 2026
1. [AutoRFP, Claude connectors for RFPs](https://autorfp.ai/blog/claude-connectors-for-rfps), read November 16, 2026