CB Insights MCP Server: Deprecated Self-Hosted Example

Overview

This repository contains a simple self-hosted example of a Model Context Protocol (MCP) server that proxies requests to the CB Insights ChatCBI LLM. It demonstrates how to expose a ChatCBI tool to MCP-enabled agents and clients so they can send messages, receive responses, and continue conversations via a chat identifier.

Important: this example is deprecated as of January 2026. CB Insights operates a fully supported MCP server at https://mcp.cbinsights.com. New integrations should use the hosted service or consult CB Insights for guidance. The self-hosted example remains useful as a reference for understanding MCP tool wiring, local testing, and how ChatCBI responses are structured.

Features

• Minimal MCP server implementation for exposing a ChatCBI tool to agents
• Pass-through proxying to the CB Insights ChatCBI API using OAuth credentials
• Conversation continuation via an optional chatID
• Local development support using the MCP inspector and uv (Astral/uv)
• Example configuration for integrating with clients like Claude Desktop

Installation / Configuration

Prerequisites:

• Python and uv (Astral) CLI installed. See https://docs.astral.sh/uv/getting-started/installation/
• MCP tooling for local inspector testing (mcp CLI)

Clone the repository and configure environment variables:

git clone https://github.com/cbinsights/cbi-mcp-server.git
cd cbi-mcp-server

Create a .env file with your CB Insights OAuth credentials and optional overrides:

# .env
CBI_CLIENT_ID=your_client_id_here
CBI_CLIENT_SECRET=your_client_secret_here
CBI_MCP_TIMEOUT=120
CBI_MCP_PORT=8000

Environment variables summary:

Run the server with uv:

uv --directory /path/to/cbi-mcp-server run server.py

Or, if using the configuration shown in the Claude Desktop example (see Use Cases):

{
  "command": "/path/to/.local/bin/uv",
  "args": [
    "--directory",
    "/path/to/cloned/cbi-mcp-server",
    "run",
    "server.py"
  ]
}

Available Tools

This MCP server exposes one primary tool: ChatCBI.

ChatCBI — send a message and receive a ChatCBI response

• Input JSON:
  • message (string) — text to send to ChatCBI
  • chatID (string, optional) — ID for continuing a prior conversation
• Typical response JSON:

{
  "chatID": "abc123",
  "message": "Here is the ChatCBI reply...",
  "relatedContent": [ /* references */ ],
  "sources": [ /* source citations */ ],
  "suggestions": [ "Follow-up question 1", "Follow-up question 2" ],
  "title": "Conversation title"
}

• Behavior: If you provide a chatID, the server will forward it to ChatCBI and the conversation will continue. If not provided, a new chat session will be started and a chatID will be returned.

Example HTTP request (local server proxy):

curl -X POST "http://localhost:8000/tools/chatcbi" \
  -H "Content-Type: application/json" \
  -d '{"message":"Tell me about the latest funding trends in fintech."}'

Example response (truncated):

{
  "chatID": "f2e9c8",
  "message": "Fintech funding has increased in Q4 driven by ...",
  "relatedContent": [],
  "sources": ["CBI-Report-2026-01"],
  "suggestions": ["Show me top-funded fintech companies"]
}

(Endpoint paths may vary depending on MCP wiring; inspect the running server via the MCP inspector to confirm.)

Use Cases

• Agent-driven Q&A: An AI agent can call the ChatCBI tool to field domain-specific questions and receive structured responses with sources and suggested follow-ups.
• Conversation continuity: Keep multi-step interactions coherent by saving chatID and passing it back in subsequent requests to continue the same ChatCBI session.
• Local testing and development: Run this example server locally to validate agent behavior and tool wiring before migrating to the hosted MCP server.
• Integration with desktop clients: Configure clients (e.g., Claude Desktop) to run a local MCP server and expose ChatCBI to the client for private or experimental workflows.

Claude Desktop configuration example:

mcp install server.py

This adds a configuration entry telling Claude Desktop how to start the local uv process that runs the example server.

Debugging & Development

• Use the MCP inspector for interactive testing and debugging:

mcp dev server.py

• Watch logs for OAuth token exchange and outbound API calls to ChatCBI.
• Adjust CBI_MCP_TIMEOUT if the upstream ChatCBI API responses take longer than the default.

Deprecation & Migration

This repository is deprecated as of January 2026. For production use, migration is recommended to the hosted, supported MCP server at:

• https://mcp.cbinsights.com

The hosted server provides support, maintenance, and guaranteed availability. Use the self-hosted example only for learning, local experimentation, or as a reference implementation.

Resources

• Repository (reference): https://github.com/cbinsights/cbi-mcp-server
• CB Insights ChatCBI API docs: https://api-docs.cbinsights.com/portal/docs/CBI-API/chatcbi
• CB Insights API authentication: https://api-docs.cbinsights.com/portal/docs/CBI-API/Authentication
• MCP inspector docs: https://modelcontextprotocol.io/docs/tools/inspector#getting-started