Deprecation notice: This repository, serving as an example implementation of a simple self-hosted MCP server that acts as a pass-through to the CB Insights API, is deprecated as of January 2026. CB Insights is pleased to offer a fully supported MCP server at mcp.cbinsights.com, which can be invoked programmatically or through clients like ChatGPT, Claude, and Microsoft Copilot.
The CBI MCP Server provides an interface for developers to interact with CB Insights ChatCBI LLM through AI Agents.
Send a message from an agent to ChatCBI and return the response.
- Input parameters:
message: The content of your message to ChatCBIchatID(optional): A unique identifier for the chat session, obtained from a previous response. If included, the conversation is continued. Otherwise, a new conversation is started.
- Returns JSON object containing the following fields:
chatID: Identifies the conversation. If chatID was provided in the request, this will be the same.message: ChatCBI response to the message.relatedContent: List of related references.sources: List of sources used to generate the response.suggestions: List of suggested follow-up questions.title: Title of the chat
- For more information see Using ChatCBI
The CBI MCP Server uses uv to manage the project.
Environment variables are set via the .env file:
CBI_CLIENT_ID&CBI_CLIENT_SECRETOAuth Client ID and Secret- see CB Insights API Authentication for usage and how to obtain them
CBI_MCP_TIMEOUT(default: 120)CBI_MCP_PORT(default: 8000)
Update the claude_desktop_config.json file using the following command:
mcp install server.pyThis will add the following configuration:
{
"mcpServers": {
"cbi-mcp-server": {
"command": "/path/to/.local/bin/uv",
"args": [
"--directory",
"/path/to/cloned/cbi-mcp-server",
"run",
"server.py"
]
}
}
}The inspector can be used to test/debug your server.
mcp dev server.py