Claude Code MCP Bridge

Run Claude Code as a one-shot MCP tool so other agents and editors can delegate coding tasks to it. An agent-in-agent orchestration bridge.

Prerequisites

Claude Code CLI (installed and authenticated), Node.js v20+, MCP-compatible client

Installation

Use the upstream install or setup path that matches your environment:

• npm install -g @anthropic-ai/claude-code
• npm test
• npm run test:unit
• npm run test:e2e

Requirements and caveats from upstream:

• Node.js v20 or later (Use fnm or nvm to install)
• Before the MCP server can use the default bypassPermissions mode, you must first run the Claude CLI manually once with the --dangerously-skip-permissions flag, login and accept the terms.
• "Generate a Python script to parse CSV data and output JSON."

Basic usage or getting-started notes:

• Run Claude Code with all permissions bypassed by default (using --dangerously-skip-permissions)

• This server is a thin MCP wrapper around the local Claude Code CLI. By default it preserves the historic behavior and starts Claude Code with --dangerously-skip-permissions. Set the tool's permissionMode argument to d...

• The wrapper cannot approve prompts that belong to a parent MCP client, bypass macOS privacy prompts, or make another Claude Code session inherit its settings. If the calling client stalls while waiting for permission...

• Source: https://github.com/steipete/claude-code-mcp

• Extracted from upstream docs: https://raw.githubusercontent.com/steipete/claude-code-mcp/HEAD/README.md

Documentation

• https://github.com/steipete/claude-code-mcp#readme

Source

• Agent Skill Exchange