Claude Code MCP Server

The Claude Code MCP (Model Context Protocol) server facilitates the execution of Claude Code in one-shot mode while bypassing permission checks automatically. This solution effectively addresses the limitations often encountered with Cursor, specifically when attempting to apply smaller edits using the edit_file command. The server leverages the magic_file tool, enhancing editing efficiency and performance.

Overview

This MCP server serves as a powerful interface for Language Learning Models (LLMs) to interact seamlessly with Claude Code, enabling functionalities that include:

• Running Claude Code with all permissions bypassed using the --dangerously-skip-permissions flag.
• Executing prompts without interruptions due to permission checks.
• Direct file editing capabilities.
• Default enabling of specific tools for enhanced functionality.

Prerequisites

To utilize this server, ensure the following requirements are met:

• Node.js version 16 or later.
• TypeScript installed (for development purposes).
• Claude CLI must be installed and operational. Verification can be conducted by executing the command /doctor, which installs or updates the CLI to the path ~/.claude/local/claude, used by this server by default.

Environment Variables

Customization of server behavior can be achieved by configuring the following environment variables:

• CLAUDE_CLI_PATH: Set an absolute path to the Claude CLI executable (optional). If specified, this path will be prioritized over default settings.
• MCP_CLAUDE_DEBUG: Enable verbose debug logging from the server to standard error by setting this variable to true (e.g., MCP_CLAUDE_DEBUG=true ./start.sh).

Connecting to Client Applications

macOS

1. Locate or create the MCP configuration file at:

   ~/.cursor/mcp.json
   

2. Add the following configuration:

   {
     "mcpServers": {
       "claude_code": {
         "type": "stdio",
         "command": "/absolute/path/to/claude-mcp-server/start.sh",
         "args": []
       }
     }
   }
   

Windows & Linux

Follow similar steps to configure the MCP server in the respective paths according to the operating system instructions provided in the documentation.

Usage

To interact with the tools provided by this server, invoke the code tool using the following format:

{
  "tool_name": "code",
  "params": {
    "prompt": "Your prompt to Claude Code here",
    "options": {
      "tools": ["Bash", "Read", "Write"]
    }
  }
}

Best Practices for Prompting

The effectiveness of the server mounts significantly on the quality of prompts. For complex tasks, detail, clarity, and structure within prompts lead to superior outcomes. For instance, breaking down multifaceted inquiries into step-by-step sequences enhances Claude's capacity to execute tasks accurately.

Example Usage: Complex Multi-step Task

The following JSON prompt format outlines how to instruct Claude on multi-step operations:

Your work folder is /path/to/project

1. Identify files to process.
2. Execute tasks specific to identified files.
3. Write outputs to designated locations.

Server Tool Description

The server incorporates a singular tool:

• Tool Name: code
  • Description: Executes a provided prompt directly via the Claude CLI, bypassing all permission checks. This tool is designed for complex workflows, facilitating code generation, analysis, and significant file system operations, including Git workflows and automated file modifications.
  • Parameters:
    • prompt (required): The full prompt directed to Claude Code.
    • options.tools (optional): Specifies any individual tools required.

Example Screenshots

Demonstrations of server capabilities can be found in the coordinated screenshots:

Troubleshooting

Common issues may arise during installation or operation, and users are encouraged to check the logs for errors and ensure proper execution permissions are set for scripts and binaries.