MCP server for Claude Code as agentic tool
Top 49.5% on sourcepulse
This project provides an MCP (Model Context Protocol) server that enables LLMs to interact with the Claude Code CLI, bypassing permission interruptions for more direct and capable coding assistance. It's designed for developers and AI agents seeking to automate complex, multi-step coding tasks, file operations, and system commands with Claude's advanced capabilities.
How It Works
The server exposes a single claude_code
tool that executes prompts directly via the Claude Code CLI, crucially using the --dangerously-skip-permissions
flag. This allows for uninterrupted execution of any prompt, including file editing, Git operations, and terminal commands. By queuing multiple commands and offloading simpler tasks to Claude Code, it aims to reduce context switching and improve efficiency compared to other LLM agents.
Quick Start & Requirements
npx -y @steipete/claude-code-mcp@latest
.--dangerously-skip-permissions
via claude --dangerously-skip-permissions
.mcp.json
or mcp_config.json
file for your MCP client.Highlighted Details
CLAUDE_CLI_NAME
environment variable.Maintenance & Community
Licensing & Compatibility
Limitations & Caveats
The initial setup requires a manual, one-time execution of the Claude CLI with --dangerously-skip-permissions
. macOS users may encounter initial permission prompts that could cause the first run to fail. Debug logging (MCP_CLAUDE_DEBUG=true
) can interfere with MCP's JSON parsing.
2 months ago
Inactive