Skip to content

How to attach an MCP client

Goal

Connect an AI agent to Codefang over the Model Context Protocol (MCP) so the agent can run code analysis as a tool.

Prerequisites

  • codefang installed and on your PATH.
  • An MCP-compatible client (for example, Claude Code or another agent that launches MCP servers over stdio).
  • For the codefang_history tool, an absolute path to a local Git repository.

Steps

  1. Confirm the MCP server starts. It communicates over stdio using JSON-RPC, the standard transport for local MCP tool use:
codefang mcp

Press Ctrl-C to stop it — the agent will launch its own copy.

  1. Register Codefang as an MCP server in your client. For Claude Code, add it to ~/.claude/settings.json (user scope) or .claude/settings.json (project scope):
{
  "mcpServers": {
    "codefang": {
      "command": "codefang",
      "args": ["mcp"],
      "env": {}
    }
  }
}

  1. Restart the client so it picks up the new server. The agent then discovers three tools via the standard MCP tools/list method: codefang_analyze (static analysis on inline code), uast_parse (parse source into a UAST), and codefang_history (history analysis on a local repo).

  2. Ask the agent to run history analysis. The codefang_history tool requires an absolute repo_path; a relative path is rejected:

{
  "name": "codefang_history",
  "arguments": {
    "repo_path": "/home/user/projects/myapp",
    "analyzers": ["burndown", "devs"],
    "limit": 500
  }
}
  1. If a call misbehaves, reproduce it on the command line: the tool result equals what codefang run prints for the same input (for example codefang run -a history/devs --format json --limit 500 /home/user/projects/myapp), and the server writes its own failures to stderr, which most clients show in their MCP log.

Result

The agent lists the three Codefang tools and returns structured analysis when it calls them. A successful codefang_analyze or codefang_history call that comes back without isError: true confirms the client is attached.

See also