A Model Context Protocol (MCP) server that brings Claude Code's powerful development tools to any application. Get instant access to file operations, command execution, web tools, and task management through the standardized MCP interface.
Transform any MCP-compatible application into a powerful development environment with:
- File Operations: Read, write, edit, search, and navigate files with ease
- Command Execution: Run shell commands with persistent sessions
- Web Tools: Fetch web content and perform searches
- Jupyter Support: Work with notebooks programmatically
- Task Management: Organize and track your work
- Built-in Security: Safe operations with automatic validation
Download the latest binary for your platform:
Linux/macOS:
curl -L https://github.com/d-kuro/claude-code-mcp/releases/latest/download/claude-code-mcp-$(uname -s | tr '[:upper:]' '[:lower:]')-amd64 -o claude-code-mcp
chmod +x claude-code-mcp
Windows:
curl -L https://github.com/d-kuro/claude-code-mcp/releases/latest/download/claude-code-mcp-windows-amd64.exe -o claude-code-mcp.exe
./claude-code-mcp
That's it! All tools are now available through the MCP interface.
- Read - View file contents with optional line ranges
- Write - Create or overwrite files safely
- Edit - Make precise string replacements
- MultiEdit - Apply multiple edits atomically
- LS - List directory contents
- Glob - Find files by patterns
- Grep - Search file contents
- Bash - Execute shell commands with persistent sessions
- WebFetch - Retrieve and process web content
- WebSearch - Search the web with filtering options
- NotebookRead - Read Jupyter notebook cells
- NotebookEdit - Modify notebook content
- TodoRead/TodoWrite - Organize tasks within sessions
Add to your ~/.config/claude-desktop/mcp.json
:
{
"mcpServers": {
"claude-code-mcp": {
"command": "/path/to/claude-code-mcp"
}
}
}
Restart Claude Desktop and the tools will appear automatically.
The server works with any MCP-compatible application. Connect using:
- stdio transport (default)
- HTTP/SSE transport (planned)
All tools work immediately with sensible defaults and built-in security.
Control logging level:
export LOG_LEVEL=debug
./claude-code-mcp
- Path Validation - All file paths are validated and sanitized
- Command Safety - Dangerous commands are blocked
- Resource Limits - File sizes and timeouts are controlled
- Session Isolation - Each MCP session is independent
- Code Review: Read and analyze code across multiple files
- Refactoring: Make coordinated changes with MultiEdit
- Debugging: Execute commands and inspect outputs
- Documentation: Fetch web content and organize information
- Research: Search files and web content
- Notebook Work: Read and modify Jupyter notebooks
- File Processing: Batch operations on multiple files
- Project Planning: Organize work with todo lists
- Progress Tracking: Maintain session-based task states
Q: Which tools are available?
A: 14 of Claude Code's 16 tools. The Task and exit_plan_mode tools are not supported in MCP context.
Q: Do I need to configure anything?
A: No! Everything works out of the box with built-in security.
Q: Can I use relative paths?
A: No, all paths must be absolute for security reasons.
Q: How do I see debug information?
A: Set LOG_LEVEL=debug
before running the server.
MIT License - see LICENSE file for details.