Salespeak Company Context is the connector to the GTM Context Layer for anyone at your company who writes with an AI assistant. The assistant asks it a question about your company: product, pricing, integrations, security, customers, positioning. The answer comes from what your company has recorded in Salespeak, with the records behind it. When two of your sources disagree, you see both versions before you write, and you choose. The connector does not decide which version is right, and it does not change what any outside AI model says about you.
https://platform.salespeak.ai/company/mcpOn this page: Connect from your client · Tools · What an answer contains · Example: our own site · Requirements · Permissions and data · Disconnect · FAQ
Connect from your client
Each page has the exact steps, a first question to try, and how to disconnect.
- Claude and Claude Code
- ChatGPT
- Cursor
- Grok Bot
- HubSpot workflows and Breeze agents, through the HubSpot app GTM Context Layer by Salespeak. It reaches the same knowledge with its own two tools, not the four below.
Any other MCP client that supports remote servers with OAuth works the same way: add the URL above and sign in with your Salespeak account. Sign-in is OAuth 2.1 with dynamic client registration, so there is no client ID, secret or API key to enter.
Tools
| Tool | Reads or writes | What it does |
|---|---|---|
ask_company_context | Reads | Answers one question about your company from its recorded knowledge, with the records behind the answer and any conflict between your sources. |
list_knowledge_conflicts | Reads | Lists where your company’s own sources disagree, most severe first. |
quote_conflict_passages | Reads | Used by the conflict card: the sentence in each record that states the disputed fact. |
flag_knowledge_conflict | Writes a flag | Tells your company’s admins that a conflict needs a decision. It changes no answer and deletes nothing. |
What an answer contains
- A status: answered, partial, or
not_in_kbwhen your company has no recorded answer. With no recorded answer, the assistant says so. - The answer, written from your knowledge bank.
- The records behind it. Each has an excerpt, its source URL and type, an authority label, and when it was last updated. Records come from the retrieval step, not from the model. They show where an answer came from; they do not certify it.
- Gaps: what the records do not say.
- Conflicts between your sources on this question, and a count of the other open conflicts in your knowledge.
In clients that show cards, such as Claude and ChatGPT, a conflict draws a card with each version, its source and its date. Use for this answer continues with one version and saves nothing. Tell an admin this needs a decision flags the conflict.
Example: our own site disagreed with itself
On October 16, 2026 we ran list_knowledge_conflicts on Salespeak’s own
knowledge, which is built from salespeak.ai. It returned two disagreements. Both were on
public pages, and both were ours. The passages below are verbatim from that result.
Contract terms (critical)
| Source | What it said |
|---|---|
| /about/ | “Contracts are month-to-month and cancel anytime.” |
| /pricing/ | “Custom pricing and annual invoice billing … Custom annual contracts” |
Meeting booking (high)
| Source | What it said |
|---|---|
| /website-inbound-agent/ | “It does not book through a calendar API, and it does not create deals…” |
| /alternatives/qualified/ | Asked “Does Salespeak support meeting booking and lead routing?”: “Yes, with full conversation context synced to the CRM in real time.” |
The pricing page was right: self-serve plans are monthly, Enterprise is a custom annual contract. The inbound agent page was right too: the agent offers your booking link and does not book through a calendar API. We corrected /about/ and /alternatives/qualified/ the same day.
The scan has a limit worth knowing. It compares the records in your knowledge bank, so it only sees what was ingested. A page-by-page read of salespeak.ai with Same Page, our free Grok Bot, found eight contradictions on the same site, including both of these.
Requirements
- A Salespeak account at your company. The connector reads only the knowledge of the company that account belongs to.
- For conflicts: an admin has run the knowledge scan once (Ground Truth, then Trust). Until then the connector answers and says that no check has run.
- An AI client that supports remote MCP servers with OAuth.
Permissions and data
- What it can reach: your company’s knowledge in Salespeak, read-only, plus the ability to store a conflict flag. Your sign-in decides which company.
- What Salespeak receives: the question your assistant asks on your behalf, not your conversation, files or history. Salespeak does not store the question or the answer.
- What a flag stores: the conflict, your email address, the version you preferred if you chose one, any note, and the time. Your admins see it in Salespeak.
Full detail is in the Company Context privacy policy.
Disconnect
Disconnect in your AI client. Each client page has the exact steps. Salespeak stops receiving questions from that client. Nothing in your company knowledge changes, and flags you sent stay for your admins.
FAQ
Can the connector change our company knowledge?
No. Three of its four tools only read. The fourth, flag_knowledge_conflict, stores a flag for your admins: the conflict, your email address, the version you preferred and any note. It changes no answer. An admin settles the conflict in the Salespeak app, and every later answer follows that decision.
Does this control what ChatGPT or Claude say about our company in public?
No. The connector answers the people at your company who sign in with a Salespeak account, inside their own assistant. It gives that assistant your recorded facts and shows where they disagree. It does not change what any AI model says to anyone else.
What happens when our knowledge has no answer?
The result says so, with the status not_in_kb, and the assistant is told not to invent one. The bundled skill for Cursor and Grok Bot says the same: do not fill the gap from general knowledge or the web.
Why does the assistant say our knowledge was never checked for conflicts?
Conflicts come from a knowledge scan that an admin runs in Salespeak (Ground Truth, then Trust). It takes about half an hour. Until it has run once, the connector still answers and says that no check has run.
Are the records behind an answer proof that the answer is right?
They show where the answer came from: each record has its source URL, its type, an authority label and the date it was last updated. A record from your crawled website is labeled discovered, not approved. Records let you trace an answer; they do not certify it.
Can an agent use it with no one signed in?
Not this endpoint. Company Context needs a person to sign in. For an agent that runs on its own, Salespeak has a separate read-only endpoint with one tool, which accepts an API key created in the Salespeak app under Publish, then API keys.
Is this the same as the Salespeak MCP server for analytics?
No. The analytics server is per workspace and answers questions about conversations, AI traffic and knowledge gaps. Company Context answers questions about your company itself. They are separate URLs.
Need help? Write to support@salespeak.ai, or see Company Context support. All other integrations are on the integrations page.