Summary

Coding Tools MCP provides a model-neutral runtime server (MCP server) designed to equip any AI agent with the ability to interact with local coding primitives. It addresses the need for AI agents to perform actions such as inspecting code, applying structured patches, running tests, and executing shell commands directly within a local development environment. This empowers AI agents by giving them direct, programmatic access to a workspace, abstracting away the underlying execution details.

How It Works

The core approach is an MCP server that exposes a defined set of local coding primitives via an HTTP endpoint or standard input/output. It is explicitly designed to be model-agnostic, meaning AI clients connect to this server to access tools like file I/O, git operations, and command execution, rather than the server being tied to a specific AI model. This separation of concerns allows flexibility in AI agent design and model choice, focusing solely on providing a secure and controlled execution environment for coding tasks.

Quick Start & Requirements

• Primary Install: curl -fsSL https://raw.githubusercontent.com/xyTom/coding-tools-mcp/main/scripts/install.sh | bash
• Run without Persistent Install: uvx coding-tools-mcp --workspace .
• Prerequisites: Python. Optional image extra requires pip install -e ".[image]".
• Links: Quickstart commands are detailed in the README. Official documentation includes docs/profile-v0.1.md, docs/remote-mcp.md, and SECURITY.md.

Highlighted Details

• Exposes core coding primitives: text inspection, file operations, structured patches, command execution, stdin interaction, and git status/diff.
• Model-neutral architecture, separating AI agent logic from the execution runtime.
• Granular security controls via exec_command permission modes (safe, trusted, dangerous) to manage execution risks.
• Supports remote access via tunnels (cloudflared, ngrok, devtunnel) with flexible authentication options.

Maintenance & Community

No specific details regarding notable contributors, sponsorships, partnerships, or community channels (like Discord/Slack) were found in the provided README.

Licensing & Compatibility

• License: Apache License 2.0.
• Compatibility: The Apache 2.0 license is generally permissive for commercial use and integration into closed-source projects, provided attribution and notice requirements are met.

Limitations & Caveats

This project explicitly omits AI agent-centric features such as memory management, web search, model routing, or subagent orchestration, leaving these responsibilities to the client. It is not a complete OS or container sandbox; security relies on configurable permission modes and host OS features like Landlock. On systems without Landlock, external sandboxing is recommended for untrusted commands. The dangerous permission mode carries significant security risks and should be used with extreme caution. The repository does not claim SWE-bench leaderboard results.