claude-code

The claude-code-mcp MCP server wraps the Claude Code CLI and exposes a set of tooling actions that you can request via JSON over stdio. It accepts requests to run tools such as explain_code, review_code, fix_code, edit_code, test_code, simulate_command, and your_own_query. Internally it Base64-encodes input to reliably handle newlines and special characters, constructs a command from the requested tool, executes it with Node.js child_process.spawn, and returns the results in JSON. To use it, run the server through npx (or your preferred installation method) and connect a client that sends the appropriate tool name and input payload; the server will respond with the execution output and any results or errors.

Prerequisites:

• Node.js (>= v18 recommended)
• npm (or yarn)
• Claude Code command installed and authenticated

Installation options:

1. Using npx (recommended for quick use)

npx @kunihiros/claude-code-mcp

2. Global installation

npm install -g claude-code-mcp

Then, run the server as a command:

claude-code-mcp

3. Local installation (development)

git clone https://github.com/KunihiroS/claude-code-mcp.git
cd claude-code-mcp/claude-code-server
npm install
npm run build

You can then run the built script directly:

node build/index.js

Environment variables can be provided via MCP Host settings (recommended for npx), a .env file, or a global ~/.claude-code-mcp.env file. Required CLAUDE_BIN should point to your Claude CLI executable. LOG_LEVEL is optional (defaults to info). The server writes to claude-code-mcp.log with a fallback order if writing to the project root fails. Ensure Claude Code CLI is installed and authenticated on the host where you run the MCP server. If you use the npx method, you may configure env vars in the MCP host configuration payload as shown in the README example.