Answer · for AI agents and their humans
How to Source Venture Deals with Claude or Cursor
Use Claude Desktop, Claude Code, or Cursor as a deal-sourcing assistant. Install the GitDealFlow MCP server (free, no API key) and ask the AI for trending technical startups, sector signals, and engineering acceleration data.
Most VC research tools were not designed for the AI-assistant era. They have dashboards, exports, and APIs — none of which compose naturally with how investors increasingly do work in Claude Desktop or Cursor. The Model Context Protocol (MCP) closes that gap. With one config-file edit, an AI assistant can call structured data tools at runtime.
Install in Claude Desktop. Open ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on Windows. Add:
``json
{
"mcpServers": {
"gitdealflow": {
"command": "npx",
"args": ["-y", "@gitdealflow/mcp-signal"]
}
}
}
``
Restart Claude Desktop. The six tools (get_trending_startups, search_startups_by_sector, get_startup_signal, get_signals_summary, get_scout_receipts, get_methodology) appear automatically.
Install in Cursor. Open Cursor Settings → Tools → MCP, paste the same JSON, save, restart Cursor. The AI in the agent panel can now call the tools.
Install in Claude Code. Edit .claude/mcp.json in your project root with the same JSON. The CLI assistant has access to the tools immediately.
Sample workflow — weekly digest review. Open Claude Desktop on Monday. Ask: 'What are this week's top 5 trending technical startups?' Claude calls get_trending_startups and returns the list. Follow up: 'Tell me more about the AI/ML ones.' Claude filters down. Follow up: 'How does $org's commit velocity compare to the AI/ML cluster median?' Claude calls get_startup_signal and returns a comparison. The whole conversation is conversational; the data is live.
Sample workflow — live diligence on inbound. A founder pitches you. Open Cursor and ask: 'Pull engineering metrics for $founder_org and tell me how it compares to our sector benchmark.' Cursor calls the tools and returns a structured answer in 15 seconds. Skip the manual GitHub dashboard work.
Sample workflow — founder taste check. Use get_scout_receipts to grade a founder's GitHub starring history against the validated unicorn list. Ask: 'What is $username's Scout Score and what unicorns did they star pre-event?' The AI returns the score and the specific unicorns.
Why MCP beats the alternatives. A REST API works but requires the user (or an agent framework) to handle parsing, prompting, and orchestration. MCP lets the AI host handle all of that natively. The user's experience is just 'ask Claude or Cursor about VC deal flow' and get answers. The free tier covers everything; paid Insider Circle adds dashboard filtering and the full universe.
Try it now
Install the MCP server →Frequently asked questions
Do I need to know how to code to install the MCP server?
No. Installation is a one-line config-file edit (no build, no compilation, no dependencies to manage). The server runs locally via npx; no infrastructure required.
Can I share the MCP install with my team?
Yes. Each team member adds the same config to their own Claude or Cursor. The MCP server runs locally per-host; there is no per-seat licensing or central admin needed.
Will Claude or Cursor know to use the tools without prompting?
Mostly yes — once the tools are registered, the AI host shows them in the toolbox and uses them when the question matches. For best results, mention the data type explicitly ('GitHub engineering acceleration', 'trending technical startups') in your first question.
Does the AI host send my data anywhere?
The MCP server itself runs locally and only makes outbound calls to the GitDealFlow public dataset endpoint. The AI host (Claude, Cursor) handles user prompts according to its own privacy policy. No GitDealFlow-specific data leakage beyond the public-dataset queries.