CB Insights MCP Server

The CB Insights MCP server provides a Model Context Protocol interface to ChatCBI, the natural-language agent that sits on top of CB Insights' market intelligence platform. Through a single conversational tool, AI agents can query the CB Insights knowledge base for private company data, funding rounds, M&A activity, partnerships, investor profiles, industry analyst notes, and competitive landscapes, then receive answers with source citations, related references, and suggested follow-up questions.

CB Insights now hosts a fully supported remote server at mcp.cbinsights.com that can be invoked from any MCP-compatible client (Claude, ChatGPT, Microsoft Copilot, Cursor, custom agents). An earlier self-hosted Python reference implementation lives at github.com/cbinsights/cbi-mcp-server, but it is deprecated in favor of the hosted endpoint.

Access requires an active CB Insights subscription. Authentication uses OAuth credentials (CBI_CLIENT_ID and CBI_CLIENT_SECRET) issued through the CB Insights API. Conversation state is preserved across calls by passing back the chatID returned in earlier responses, allowing agents to drill down on a topic over multiple turns.

Tools

Prerequisites

• Active CB Insights subscription with API access
• CBI_CLIENT_ID and CBI_CLIENT_SECRET obtained from the CB Insights API Authentication documentation
• An MCP-compatible client (Claude Desktop, Cursor, ChatGPT, Copilot, or a custom client)

Option A: Hosted server (recommended)

Point your MCP client at the official hosted endpoint at https://mcp.cbinsights.com and authenticate with your CB Insights OAuth credentials.

{
  "mcpServers": {
    "cb-insights": {
      "url": "https://mcp.cbinsights.com"
    }
  }
}

Option B: Self-hosted reference server (deprecated)

The legacy Python implementation at github.com/cbinsights/cbi-mcp-server uses uv for dependency management.

1. Clone the repo and install dependencies with uv.
2. Create a .env file with:
   • CBI_CLIENT_ID (required)
   • CBI_CLIENT_SECRET (required)
   • CBI_MCP_TIMEOUT (optional, default 120)
   • CBI_MCP_PORT (optional, default 8000)
3. Install into Claude Desktop:

mcp install server.py

4. Debug with the MCP inspector:

mcp dev server.py

Note: CB Insights recommends migrating any self-hosted deployments to the hosted server at mcp.cbinsights.com.

• Pull private company profiles, headcount, and product info on a target list of startups before an outbound campaign
• Research the latest funding rounds, lead investors, and valuations in a specific industry vertical
• Build a competitive landscape map by asking ChatCBI to identify and compare key players in an emerging tech category
• Surface recent M&A activity, partnerships, and patents for a named company to brief a sales or strategy team
• Generate sourced market sizing and trend summaries that can be dropped into investment memos or board decks

• "Use CB Insights to list the top 10 best-funded vertical AI startups in financial services with their last round size and lead investor."
• "Ask ChatCBI for a competitive landscape of robotics process automation vendors and include sources."
• "Summarize the last 12 months of M&A activity in cybersecurity, then continue the same ChatCBI conversation to drill into identity security deals."
• "Get a CB Insights profile of Anthropic including funding history, key investors, and notable partnerships."
• "Find emerging climate-tech companies working on direct air capture and return CB Insights' suggested follow-up questions."

• Official, provider-maintained server backed by CB Insights' proprietary market intelligence data
• Returns citations, source references, and suggested follow-ups, which makes outputs auditable
• Hosted endpoint at mcp.cbinsights.com works with any MCP client without self-hosting
• Conversation continuity via chatID lets agents drill down across multiple turns

• Requires a paid CB Insights subscription and OAuth credentials, not usable on a free tier
• Exposes only a single high-level ChatCBI tool rather than granular endpoints for companies, deals, or investors
• The self-hosted reference implementation on GitHub is deprecated, so customization options are limited

• PitchBook for private market data (no official MCP server at this time, API integrations only)
• Crunchbase via community MCP servers wrapping the Crunchbase API
• Exa or Tavily MCP servers for broader web-based company and market research