skills.rest
reason-machines
reason-machinesOFFICIAL
Explore all skills by this author →
Website𝕏@reasonmachinesGitHub

coding-tools-mcp-server

Exposes workspace-bounded file, git, and shell command primitives to MCP clients.

7|2|Updated May 16, 2026
Search Tags:#git#command execution#MCP server#file operations#patching#coding agent#workspace security
One-click install
npx skills add https://github.com/reason-machines/mcp-skills --skill coding-tools-mcp-server-reason-machines
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: coding-tools-mcp-server
Source: https://github.com/reason-machines/mcp-skills/tree/main/skills/coding-tools-mcp-server
Command: npx skills add https://github.com/reason-machines/mcp-skills --skill coding-tools-mcp-server-reason-machines

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? AI coding agents need controlled access to local files, git history, and shell commands without risking path traversal, secret leakage, or destructive operations. This Skill configures an MCP server that enforces workspace boundaries and permission gates around those primitives. ## Core Features & Use Cases - Workspace-Bounded File Operations: Read, list, and search files with automatic exclusion of build artifacts, plus unified-diff patch application with dry-run preview. - Safe Command Execution: Run shell commands with timeouts, output caps, environment filtering, and three permission modes (safe, trusted, dangerous), including interactive stdin sessions. - Read-Only Git Inspection: Query status, diff, log, show, and blame without mutating the repository. - Use Case: Connect Claude Code or Cursor to a repository, then have the agent search for a function, preview a patch with dry_run, apply it, and verify the result with git_diff — all confined to the workspace. ## Quick Start Set up the coding-tools MCP server for my repository at /path/to/repo in safe mode and connect it to Claude Code.

Frequently Asked Questions about coding-tools-mcp-server

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I set up an MCP server for file and git access in Claude Code?▼

Add a coding-tools entry to claude_desktop_config.json with command uvx and args coding-tools-mcp --stdio --workspace /path/to/repo. The server then exposes read_file, search_text, apply_patch, and read-only git tools to the client.

How to run shell commands safely through an MCP server?▼

Use the exec_command tool with the default safe permission mode, which blocks shell expansion, network-looking commands, and destructive operations. Commands run workspace-bounded with timeouts, output caps, and sensitive environment variable filtering.

Does coding-tools-mcp work with Cursor and other MCP clients?▼

Yes, it is model-neutral and works with any MCP client via stdio or Streamable HTTP using protocol version 2025-06-18. Cursor configuration mirrors Claude Code, and a generic TOML configuration is also supported.

Why does exec_command get permission denied in safe mode?▼

Safe mode blocks network commands, shell expansion, and risky operations by default. Either switch to trusted mode for local development or call the request_permissions tool so the client can prompt the user for approval.

Can I expose the coding-tools MCP server remotely?▼

Yes, the tunnel.sh script supports cloudflared, ngrok, and Microsoft Dev Tunnel with bearer token or OAuth 2.1 authentication. The noauth mode is intended for testing only and should pair with the read-only tool profile.

What are the limitations of coding-tools-mcp as a sandbox?▼

It is not a full sandbox; Landlock filesystem confinement only works on Linux kernel 5.13 or later. For untrusted code, use external isolation such as Docker containers, VMs, or Windows Sandbox.

Related Skills

Explore More Software Engineering Skills→
archibatearchibate
6

uv

Manage Python packages, virtual environments, and MCP servers with uv.

Community
Intermediate
IntentArchitectIntentArchitect
23

address-pr-review

Resolves unresolved pull request review comments with per-thread commits and staged replies.

Official
Advanced
ArthTech-AIArthTech-AI

qa

Run full-stack QA checks with pytest, Jest, ruff, TypeScript, and Playwright.

Official
Intermediate
Heyo-ComputerHeyo-Computer

heyvm-sandbox

Manage lifecycle, execution, and networking of isolated virtual machine sandboxes.

Official
Advanced