xyTom/coding-tools-mcp

What it solves

It provides a secure, model-neutral runtime that allows AI chat applications and agents to interact with a local codebase. It transforms standard chat interfaces into coding agents by giving them a safe way to read files, search code, execute commands, and manage Git operations without requiring the user to switch to a specialized coding product.

How it works

The project implements the Model Context Protocol (MCP), serving a catalog of 18+ tools to any compatible MCP client (such as Claude Desktop, Cursor, or VS Code). It operates within a defined workspace root to prevent directory traversal attacks and uses three permission modes—safe, trusted, and dangerous—to gate risky actions like network access or destructive commands. On Linux, it leverages Landlock for kernel-level filesystem confinement.

Who it’s for

• AI users who want to turn their existing chat clients into full-fledged coding agents.
• Agent developers who need a production-grade set of file and execution tools with built-in safety boundaries rather than building them from scratch.
• Developers who want to run AI agents on their local machine or in isolated Docker sandboxes for untrusted code.

Highlights

• Model-Neutral: Works across different LLMs and vendors via the MCP standard.
• Comprehensive Toolset: Includes tools for file reading, multi-file patching, text search, command execution (including interactive REPLs), and Git integration.
• Context-Optimized: Results are summarized and paginated to reduce token usage in LLM context windows.
• Flexible Deployment: Supports stdio and HTTP transports, remote access via tunnels, and containerized sandboxing.