MCP Server

Capvia exposes a Model Context Protocol server so MCP-capable LLM clients — Claude Desktop, claude.ai, ChatGPT, and others — can read Capvia data directly. Instead of opening the app, scrolling a list, and copy-pasting into a chat window, you point your assistant at a URL and ask in plain language.

Capvia is organised as multiple MCP servers, one per data domain. Each lives at its own URL and exposes a focused set of tools. You can connect any combination of them, and they all share the same authentication.

Available servers

Each server below is a global Capvia MCP endpoint — you connect once, sign in to your Capvia account, and the tools become available to your LLM.

Accelerator Directory

Read the global accelerator database — the same catalogue that powers Discover → Accelerators in the app. The LLM can list and filter programmes by country, region, and international-friendliness, look up a specific accelerator's profile (overview, terms, application timeline, alumni), and run natural-language searches.

Useful for: shortlisting programmes, comparing deadlines and equity terms, and grounding outreach in the actual programme data rather than a screenshot.

Investor Directory

Read the global investor catalogue — VCs, corporate VCs, family offices, angels, and other investment organisations. Mirrors the data behind Discover → Investors. The LLM can search by firm name, filter by country, market region, firm type, or investor role, and pull a specific firm or individual investor's full profile.

Useful for: building target lists for a raise, doing pre-meeting research, and drafting outreach that actually references the right partner and thesis.

Investors Product (deal flow)

The investor-side product surface — your deal pipeline, AI rubric scoring, human scorecards, and investment decisions. Unlike the directories, this server is account-scoped: every request operates on the data belonging to the Capvia account you authorise.

Useful for: querying your live deal pipeline, drafting memos grounded in your own scorecards, ranking deals by rubric, and asking comparative questions across your portfolio without leaving your assistant.

Only relevant if you use Capvia's investor product (deal flow / scorecards / rubrics). Startup-side founders will rarely use this server.

How authentication works

You sign in through your browser — no API keys to generate, paste, or rotate.

When you add a Capvia MCP server to your client, it opens the Capvia sign-in page, you pick which account to authorise, and click Approve. The client handles the rest. Sessions last roughly 30 days before you're prompted to re-authorise.

Under the hood the flow uses OAuth 2.1 with PKCE under the mcp:access scope. Your client doesn't need any client-secret configuration — paste the URL and follow the prompts.

For self-hosted or scripted setups, Bearer-token auth via Authorization: Bearer cap_... is also supported. Generate a token from your account settings.

Connecting Claude Desktop and claude.ai

1. In Claude Desktop (or at claude.ai), open Settings → Connectors, scroll to the bottom, and click Add custom connector.
2. Paste one of the server URLs above (e.g. https://api.capvia.ai/mcp/accelerators) and click Add.
3. Claude opens a Capvia sign-in window. Sign in (Google or email/password), choose the account to authorise, and click Approve.
4. Back in Claude, the Capvia tools appear in the tool picker. Repeat for each server you want to connect.

Try a prompt once it's connected:

Find five accelerators in EMEA that accept international applicants and rank them by alumni quality.

Claude will call the accelerator tools and answer from the live catalogue.

Custom connectors are available across Claude's Free (one connector), Pro, Max, Team, and Enterprise plans.

Connecting ChatGPT

Custom MCP connectors in ChatGPT require Developer Mode, which is available on Business, Enterprise, and Edu plans.

1. Enable Developer Mode under Settings → Apps & Connectors → Advanced settings.
2. Go to Settings → Connectors → Create, paste a Capvia server URL, give it a name, and submit.
3. ChatGPT will open a Capvia sign-in window — sign in, choose your account, and click Approve.
4. The Capvia tools now appear in the tool picker.

Try:

Which UK seed-stage VCs in my target list have invested in fintech in the last 18 months?

ChatGPT will pull from the Investor Directory and ground the answer in real data.

OpenAI is renaming "Connectors" to "Apps", so menu labels may have shifted — the flow is unchanged.

What makes this useful

The point of the MCP server is to collapse the open-the-app-and-summarise loop into a single sentence. A connected LLM can answer questions like:

• "Shortlist accelerators in North America taking pre-seed climate-tech, and write me a one-paragraph cover note for the top three."
• "Which partners at the firms in my target list have posted publicly about AI infrastructure in the last six months?"
• "Compare the investment terms across the three accelerators I'm down to and tell me which is best for a UK pre-seed company."
• "Summarise my current deal pipeline by stage and flag any deals that haven't moved in 14 days." (Investors product only)

Because the LLM is calling structured tools against your live Capvia data, the answers are grounded — not a scrape, not a screenshot, not last week's CSV.

Revoking access

Removing the connector from your MCP client ends the session. To revoke access immediately from the Capvia side, contact support — self-serve connection management is on the roadmap.

Related

• Introduction — what Capvia tracks and where the MCP data comes from
• Investor Database — the same data, browsable in the app
• Accelerator Database — the same data, browsable in the app