MCP Server

AFA ships an MCP (Model Context Protocol) server that exposes analysis and enhancement as tools for AI-powered IDEs.

Start the server

afa serve --transport stdio

The server communicates over stdio, which is the standard transport for MCP in IDE integrations.

Add AFA as an MCP server in your Claude Code configuration:

{
  "mcpServers": {
    "afa": {
      "command": "afa",
      "args": ["serve", "--transport", "stdio"]
    }
  }
}

Once configured, Claude Code can invoke AFA tools directly during conversations.

Add to your Cursor MCP settings (.cursor/mcp.json):

{
  "mcpServers": {
    "afa": {
      "command": "afa",
      "args": ["serve", "--transport", "stdio"]
    }
  }
}

Tool	Description
`afa_analyze`	Analyze a code snippet or file for security, performance, maintainability, and documentation issues
`afa_enhance`	Generate an enhancement candidate and evaluate it through all nine gates
`afa_gate_check`	Evaluate a set of metrics against the gate engine

Resource	Description
`afa://status`	Current AFA configuration and status
`afa://history`	Recent audit trail entries

Once configured, you can ask Claude Code to use AFA:

"Analyze this function with AFA and tell me what it found."

"Run an AFA enhancement on utils.py in dry-run mode."

"Check if these metrics would pass AFA's gates."

The IDE agent will invoke the appropriate MCP tool and present the results inline.

AFA must be installed and available on your PATH
An LLM API key must be set (ANTHROPIC_API_KEY, OPENAI_API_KEY, or GOOGLE_API_KEY)
Developer tier or above for full MCP server access (Community tier does not include mcp_server feature)

For non-IDE integrations, you can also run the MCP server over HTTP:

afa serve --transport http --port 8080

This starts a FastAPI server with both REST API endpoints and MCP protocol support. See REST API for endpoint documentation.