ContextDigger MCP Server

A lightweight Model Context Protocol server that turns ContextDigger CLI commands into MCP tools for clients like Claude Desktop.

What it is

The contextdigger-mcp package is a small bridge between MCP compatible tools and your locally installed ContextDigger CLI. It does not replace the CLI and it does not run as a remote service. 

AI tool (for example Claude Desktop)
        ↓  Model Context Protocol
contextdigger-mcp  (this package)
        ↓  subprocess calls
contextdigger CLI  (separate installation)
        ↓
your .cdg/ data on disk

Every MCP tool in the server maps to a real CLI command, so behavior stays consistent with what you see in the terminal.

Install and connect to an MCP client

  1. Install the CLI: pip install contextdigger
  2. Install the MCP server: pip install contextdigger-mcp
  3. Edit your MCP client configuration (for example Claude Desktop) and add: 
    {
  "mcpServers": {
    "contextdigger": {
      "command": "contextdigger-mcp"
    }
  }
}
  4. Restart Claude Desktop so it discovers the new MCP server.

If you use a virtual environment, point the command field at the full path to the contextdigger-mcp binary. Other MCP compatible tools follow the same idea: declare a server named contextdigger whose command runs the MCP executable on your machine.

What you get through MCP

The server exposes 83 tools that mirror the CLI. Some examples:

  • Navigation: list areas, dig into an area, go back and forward, show status and history.
  • Bookmarks and notes: create and jump to bookmarks, add notes, generate a quick wiki.
  • Governance: check budgets, split large areas, create sub areas, inspect provenance.
  • Analytics: insights, hotspots, trends, impact analysis, and reports.

From the perspective of the AI tool, ContextDigger becomes a native MCP server. From your perspective, it is still the same governed bundles and `.cdg/` data on disk.

Why use MCP instead of copy and paste

  • No manual pasting of bundles for routine discovery or navigation tasks.
  • Structured outputs and parameters instead of free form chat prompts.
  • One place to maintain the mapping between CLI commands and assistant actions.

You can still copy governed bundles into any assistant by hand. MCP is an additional surface for tools that know how to speak the protocol.