What is MCP?

Model Context Protocol (MCP) is an open standard introduced by Anthropic that enables AI assistants to connect with external tools and data sources in a standardized way. Think of it like USB for AI - one protocol that works everywhere.

Learn more about MCP at the official MCP introduction guide.

Code Pathfinder implements an MCP server that exposes your codebase's structure, call graphs, and relationships to AI assistants through 6 powerful tools.

Prerequisites

Before setting up the MCP server, you need:

1. Code Pathfinder installed on your system
2. An MCP-compatible AI assistant (Claude Code, Open Code, Codex, etc.)
3. A Python project to analyze (currently Python only, more languages coming soon)

Installation

Install Code Pathfinder

👉 Follow the Quickstart Guide for detailed installation instructions.

The quickstart guide covers all installation methods:

• Homebrew (macOS/Linux)
• pip (Python 3.8+)
• Docker
• From Source

Setting Up with AI Assistants

Code Pathfinder MCP server works seamlessly with Claude Code, Open Code, Codex, and other MCP-compatible AI assistants.

Step 1: Locate Your Configuration File

Your AI assistant stores its MCP server configuration in a JSON file. The location depends on your tool:

Claude Code (macOS/Linux):

~/.config/claude/mcp_config.json

Open Code / VS Code with MCP:

~/.vscode/mcp_servers.json

Codex:

~/.config/openai/codex/config.toml

Step 2: Add MCP Server Configuration

Open the configuration file in your text editor and add the Code Pathfinder MCP server.

For Claude Code / Open Code / VS Code:

{
  "mcpServers": {
"code-pathfinder": {
  "command": "pathfinder",
  "args": [
    "serve",
    "--project",
    "/absolute/path/to/your/project"
  ]
}
  }
}

For Codex:

[mcp_servers.code-pathfinder]
command = "pathfinder"
args = ["serve", "--project", "/absolute/path/to/your/project"]

Important:

• Use absolute paths for the --project argument
• Replace /absolute/path/to/your/project with your actual project directory
• Do NOT use ~ for home directory - expand it to the full path

Multiple Projects

You can configure multiple projects by adding more MCP servers.

For Claude Code / Open Code / VS Code:

{
  "mcpServers": {
"project-backend": {
  "command": "pathfinder",
  "args": ["serve", "--project", "/Users/you/code/backend"]
},
"project-frontend": {
  "command": "pathfinder",
  "args": ["serve", "--project", "/Users/you/code/frontend"]
}
  }
}

For Codex:

[mcp_servers.project-backend]
command = "pathfinder"
args = ["serve", "--project", "/Users/you/code/backend"]

[mcp_servers.project-frontend]
command = "pathfinder"
args = ["serve", "--project", "/Users/you/code/frontend"]

Custom Python Version

Override automatic Python version detection.

For Claude Code / Open Code / VS Code:

{
  "mcpServers": {
"code-pathfinder": {
  "command": "pathfinder",
  "args": [
    "serve",
    "--project", "/path/to/project",
    "--python-version", "3.11"
  ]
}
  }
}

For Codex:

[mcp_servers.code-pathfinder]
command = "pathfinder"
args = ["serve", "--project", "/path/to/project", "--python-version", "3.11"]

Step 3: Test Your Setup

Try these example queries in your AI assistant:

Example Query:

Find all functions that call process_payment

Expected Response:

The process_payment function is called by:
- checkout_handler (line 45 in handlers/checkout.py)
- subscription_renewal (line 89 in services/billing.py)
- refund_processor (line 23 in handlers/refunds.py)

Example Query:

Show me what functions validate_user calls

Expected Response:

The validate_user function calls:
- check_email_format (line 12 in utils/validators.py)
- verify_password_strength (line 34 in utils/validators.py)
- query_user_database (line 67 in database/users.py)

If your AI assistant responds with similar code intelligence from your project, everything is working!

HTTP Transport (Optional)

For remote access or custom integrations, you can run the MCP server in HTTP mode:

$ pathfinder serve --project /path/to/project --http --address :8080

This starts an HTTP server on port 8080 that accepts MCP requests over HTTP instead of stdio.

Use cases for HTTP mode:

• Team collaboration (shared code analysis server)
• CI/CD integrations
• Custom web-based tools
• Remote development environments

$ which pathfinder
# Should print: /usr/local/bin/pathfinder (or similar)

ls /path/to/project

$ pathfinder serve --project /path/to/project
# Watch for errors in terminal output

{
  "mcpServers": {
    "code-pathfinder": {
      "command": "pathfinder",
      "args": [
        "serve",
        "--project", "/path/to/your/project",
        "--python-version", "3.11"
      ]
    }
  }
}