Crunchbase MCP MCP server icon

Crunchbase MCP

by Crunchbase MCP

HIPAA CompliantSOC2 ReadyISO 27001 Ready
Productivity16 tools

Research private companies, funding rounds, investors and people from an AI agent. Anthropic lists 16 tools; Crunchbase documents 14 and its support article names 13. Three list tools write. A metering tool reports remaining quota.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Connect Crunchbase MCP via MCP

https://mcp.crunchbase.com/

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Crunchbase MCP Tools & Capabilities (16)

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Three tools write, and Crunchbase names them itself. cb_list_create and cb_list_add_entities modify data in your workspace, and Crunchbase's advice is to "keep on manual approval until you trust the workflow". A created list is immediately visible to your whole team, and no tool deletes a list or removes an entity from one — so a mistaken write is corrected in the Crunchbase UI, not by the agent.
  • Two directory tools are undocumented. cb_expert_submit_search_feedback and cb_knowledge_center appear in Anthropic's listing and in neither Crunchbase enumeration. The first is, by name, a fourth write; what it records is unpublished. We will not guess at the second.
  • No application scopes exist. The server's RFC 9728 descriptor advertised exactly one scope on 2026-08-23 — offline_access, a refresh scope that grants no capability and draws no boundary. Crunchbase's authorization server additionally advertises lists.read; that document is served at the vendor's own host root and describes Crunchbase's OAuth vocabulary generally, so we treat it as a lead rather than as this connector's grant. Either way no scope separates reading a company profile from creating a list, which is the boundary that matters here, and lists.read names a narrower verb than the three list writes actually perform.
  • No numeric MCP quota is published. Crunchbase says limits depend on "your package or agreement". The 200-calls-per-minute figure is the REST API's, and Crunchbase states MCP and API quotas are separate. cb_metering_get_state reports the real state at runtime; nothing reports it beforehand.
  • We could not read tool schemas or safety annotations. The endpoint returned HTTP 401 to an anonymous request, so no readOnlyHint or destructiveHint is published here. Parameter detail above comes from Crunchbase's Tool Reference, not from the server.
  • The contracts are silent on RAG, vectors and embeddings — zero occurrences across four instruments — while restriction (o) and §2(iii) prohibit any use that makes the data impossible to expunge. Silence is not permission, and an unpurgeable index is the clause's plain target.
  • Termination requires certified destruction in ten days, backed by a one-year audit right, with the restrictions surviving termination. Design retrieval so Crunchbase-sourced content can be deleted by provenance on day one; retrofitting that into an embedding store is materially harder.
  • Unattended automation is outside the seat model. Seats are per human sign-in; Crunchbase routes automation to its account team, and its terms of service forbid "processes that run or are activated while you are not logged into the Service". A scheduled agent holding a user's credentials is the case to get in writing.
  • Programmatic access is a separate, unpublished arrangement. Crunchbase describes headless and backend access with "their own authentication, identity, and data-access considerations", arranged through a CSM. Its tool surface is not documented, so nothing on this page describes it.
  • Data access is entitlement-shaped, so absence is not evidence of absence. Crunchbase states available data "depend[s] on your Crunchbase package, enabled products, and permissions" and that MCP grants nothing outside your agreement. A field missing from a response may exist in Crunchbase.
  • Predictions are directional and disclaimed. Funding, exit, growth and layoff forecasts are model output, which Crunchbase says should not be treated as guarantees and is not investment advice. Crunchbase separately warns its AI Content "may occasionally generate incorrect information".
  • No prompts or resources could be enumerated. Anthropic's directory records no prompt names, and the authentication gate prevented reading the server's declared capabilities.

Frequently asked questions

No. Three of the sixteen tools write. Crunchbase's own Tool Reference names cb_list_create, cb_list_add_entities and cb_expert_submit_feedback, and states plainly that the list tools modify data in your Crunchbase workspace. Nothing deletes, and no tool removes entities from a list, so the writes are additive and visible to your whole team.

Sixteen by Anthropic's count, and Crunchbase publishes two smaller numbers. Its Tool Reference documents 14 and claims to cover every tool; its support article says 13. Both vendor lists are strict subsets of the directory's, so nothing is fabricated, but two names appear only in Anthropic's listing.

Training is banned outright; ordinary inference is not, and bulk copying into a third-party AI tool is. Restriction (n) of the terms of service prohibits using Content to train models. Restriction (j) prohibits storing any significant portion or integrating it with a third-party tool, naming ChatGPT as the example.

Crunchbase never mentions vector stores or embeddings in its contracts, but one clause reaches them by function. Restriction (o) prohibits using Content so that it becomes impossible to expunge, and an embedding index is difficult to purge by source. Its MCP support article separately names persistent embeddings as a restriction category.

You destroy it within ten days and an officer signs for it. Section 11 of the Data Access Terms requires removing all copies from every computer, hard drive, network and storage medium, with officer certification. Crunchbase may audit your systems for a year afterwards, and the restrictions in Section 2 survive termination.

No. Crunchbase states seats are per user and each connection is tied to the individual who signs in, and routes unattended or shared automation to its account team instead. The terms of service separately forbid sharing an account and forbid processes that run while you are not logged in.

Yes, about founders and executives who never transacted with you. Crunchbase's data dictionary lists person fields including first, middle and last name, gender, date of birth, date of death, biography, profile photo, primary job title, and LinkedIn, Facebook and X profile links. Search covers a people collection by default.

No. Access is a purchased seat, so a call spends quota rather than money, and exceeding a limit returns an error instead of a charge. Unusually, the surface can see its own budget: cb_metering_get_state reports which limits apply and how much remains. Crunchbase publishes no numeric MCP quota.

Yes, whenever output reaches a third party. Crunchbase's licence agreement requires attributing content sourced or derived from Crunchbase, with a hyperlink that is plainly visible, sits close to the data, is visible to search engine crawlers, and carries no nofollow tag. The word derived reaches AI-generated summaries.

Sources

Use in Agentman

Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio

Server Info

Category
Productivity
Developer
Crunchbase MCP
Tools
16
Domain
mcp.crunchbase.com

Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.