Skip to content
Back to search

coding-tools-mcp

Give any AI agent the ability to code

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/xytom--coding-tools-mcp/badge.svg)](/listings/xytom--coding-tools-mcp)

HTML

<a href="/listings/xytom--coding-tools-mcp"><img src="/api/listings/xytom--coding-tools-mcp/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Run it with whichever toolchain you already have (the server is Python ≥ 3.11 from PyPI; the npm package is a thin launcher that starts it via `uv` or `pipx`):

uvx coding-tools-mcp --stdio --workspace /path/to/repo   # Python toolchain
npx coding-tools-mcp --stdio --workspace /path/to/repo   # Node toolchain

Wire it into Claude Desktop, Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, or Cline — the JSON is the same everywhere (swap `uvx` for `npx` if you prefer Node):

{
  "mcpServers": {
    "coding-tools": {
      "command": "uvx",
      "args": ["coding-tools-mcp", "--stdio", "--workspace", "/path/to/repo"]
    }
  }
}

Then ask your client: *"run the test suite and fix the first failure."*

Prefer HTTP? Drop `--stdio` and the server speaks Streamable HTTP on `http://127.0.0.1:8765/mcp`. Both protocol eras are served on either transport: MCP `2026-07-28` in full, with `tools` as the only advertised capability, and the handshake era `2025-11-25` with `2025-06-18` compatibility. Neither has sessions. A one-line installer, per-client walkthroughs, and troubleshooting live in [docs/quickstart.md](docs/quickstart.md) and [docs/mcp-client-config.md](docs/mcp-client-config.md).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 228 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Coding Tools MCP
  2. **English** | [简体中文](README.zh-CN.md)
  3. > Give any AI chat or agent a safe pair of hands on your codebase.
  4. [![PyPI](https://img.shields.io/pypi/v/coding-tools-mcp)](https://pypi.org/project/coding-tools-mcp/)
  5. [![npm](https://img.shields.io/npm/v/coding-tools-mcp)](https://www.npmjs.com/package/coding-tools-mcp)
  6. [![Python](https://img.shields.io/pypi/pyversions/coding-tools-mcp)](https://pypi.org/project/coding-tools-mcp/)
  7. [![compliance](https://github.com/xyTom/coding-tools-mcp/actions/workflows/compliance.yml/badge.svg)](https://github.com/xyTom/coding-tools-mcp/actions/workflows/compliance.yml)
  8. [![release](https://github.com/xyTom/coding-tools-mcp/actions/workflows/release.yml/badge.svg)](https://github.com/xyTom/coding-tools-mcp/actions/workflows/release.yml)
  9. [![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](LICENSE)
  10. Coding Tools MCP is a **model-neutral coding runtime** served over the
  11. [Model Context Protocol](https://modelcontextprotocol.io): file reading and
  12. search, structured multi-file patches, command execution, interactive
  13. sessions, and git — one server that any MCP client can drive. Claude Desktop,
  14. Claude Code, Codex, Cursor, Cline, VS Code, Windsurf, Gemini CLI, or an agent
  15. you build yourself gets the default catalog of 18 battle-tested tools, confined
  16. to one workspace and gated by permission modes.
  17. [![Watch the demo](https://img.youtube.com/vi/N9lQaXt1eqQ/maxresdefault.jpg)](https://youtu.be/N9lQaXt1eqQ?si=LyEwvzzQF6QjUxR0)
  18. ## Why people use it
  19. - **It turns a chat app into a coding agent.** Claude Desktop — or any MCP
  20. chat client — gets real repo access with the subscription you already have.
  21. No extra product required.
  22. - **Safety is the product, not an afterthought.** One workspace root per
  23. server. Absolute paths, `..` traversal, and symlink escapes are rejected.
  24. Permission modes gate network access, shell expansion, inline scripts, and
  25. destructive commands. On Linux, [Landlock](docs/security-boundary.md) adds
  26. kernel-level filesystem confinement.
  27. - **It is model- and vendor-neutral.** A truthfully annotated, mode-aware
  28. catalog — no profile switching, no annotation games. Swap models or clients
  29. freely; the runtime contract stays put.
  30. - **It is engineered for context windows.** Results are summarized, paginated,
  31. and capped by design; serialized tool-result bytes dropped 37%
  32. release-over-release on the deterministic dogfood workload with unchanged
  33. task completion.
  34. ## Quickstart
  35. Run it with whichever toolchain you already have (the server is Python ≥ 3.11
  36. from PyPI; the npm package is a thin launcher that starts it via `uv` or
  37. `pipx`):
  38. ```bash
  39. uvx coding-tools-mcp --stdio --workspace /path/to/repo # Python toolchain
  40. npx coding-tools-mcp --stdio --workspace /path/to/repo # Node toolchain
  41. ```
  42. Wire it into Claude Desktop, Claude Code, Codex, Cursor, VS Code, Windsurf,
  43. Gemini CLI, or Cline — the JSON is the same everywhere (swap `uvx` for `npx`
  44. if you prefer Node):
  45. ```json
  46. {
  47. "mcpServers": {
  48. "coding-tools": {
  49. "command": "uvx",
  50. "args": ["coding-tools-mcp", "--stdio", "--workspace", "/path/to/repo"]
  51. }
  52. }
  53. }
  54. ```
  55. Then ask your client: *"run the test suite and fix the first failure."*
  56. Prefer HTTP? Drop `--stdio` and the server speaks Streamable HTTP on
  57. `http://127.0.0.1:8765/mcp`. Both protocol eras are served on either
  58. transport: MCP `2026-07-28` in full, with `tools` as the only advertised
  59. capability, and the handshake era `2025-11-25` with `2025-06-18`
  60. compatibility. Neither has sessions. A one-line installer, per-client
  61. walkthroughs, and troubleshooting live in
  62. [docs/quickstart.md](docs/quickstart.md) and
  63. [docs/mcp-client-config.md](docs/mcp-client-config.md).
  64. ## Seven things to try
  65. **1. Make Claude Desktop your coding agent.** The config above is all it
  66. takes — the chat window you already pay for can now read, patch, test, and
  67. commit-review a real repository.
  68. **2. Code on your own machine from anywhere.**
  69. ```bash
  70. CODING_TOOLS_MCP_AUTH_MODE=bearer ./integrations/tunnels/tunnel.sh cloudflared /path/to/repo
  71. ```
  72. Loopback bind + authenticated HTTPS tunnel (`cloudflared`, `ngrok`, or
  73. Microsoft Dev Tunnel). Point claude.ai on your phone at
  74. `https://<tunnel-host>/mcp` and drive your home workstation from anywhere.
  75. ChatGPT and Grok connect through their connector settings the same way.
  76. Bearer tokens and OAuth 2.1 + PKCE (with RFC 7591 dynamic registration) are
  77. built in. → [docs/remote-mcp.md](docs/remote-mcp.md)
  78. **3. Let an agent loose on untrusted code — inside a disposable sandbox.**
  79. ```bash
  80. docker build -t coding-tools-mcp-sandbox:local .
  81. docker run --rm --init -it -p 8765:8765 -v "$PWD:/workspace" coding-tools-mcp-sandbox:local
  82. ```
  83. A containerized server with toolchains and caches preconfigured, safe to point
  84. at a sketchy PR and destroy afterwards. → [docs/docker.md](docs/docker.md)
  85. **4. Spin up a cloud sandbox with one MCP call.** The bundled
  86. [Cloudflare Worker control plane](infra/cloudflare/sandbox-control/README.md) exposes
  87. `start_coding_tools_sandbox` as an MCP tool: one call dispatches a GitHub
  88. Actions runner that boots the Docker sandbox and publishes it behind an
  89. authenticated Cloudflare Tunnel. Ephemeral compute, no server of your own.
  90. **5. Drive it from a GUI.**
  91. ```bash
  92. python -m pip install "coding-tools-mcp[desktop]"
  93. coding-tools-mcp-desktop
  94. ```
  95. Per-workspace profiles, server and tunnel start/stop, credential setup with
  96. clipboard helpers, live health checks. English and 简体中文.
  97. **6. Keep an interactive command alive.** `exec_command` starts a REPL or
  98. debugger under a real PTY; `write_stdin` feeds it across turns; `read_output`
  99. pages long output; `kill_command` cleans up. Long-running processes are
  100. first-class, with deadline watchdogs and bounded buffers.
  101. **7. Give your own agent production-grade hands.** Building an agent loop with
  102. the Anthropic SDK or anything else? Don't hand-roll file and exec tools —
  103. speak MCP to this server and inherit the whole safety boundary. →
  104. [docs/embedding.md](docs/embedding.md)
  105. ## The tool catalog
  106. The registry contains 19 truthfully annotated tools. The default `safe` and
  107. `trusted` modes advertise 18; `dangerous` also advertises
  108. `request_permissions`, the only mode in which that tool can grant anything.
  109. `apply_patch` and `apply_changes` are the file-mutation primitives: both are
  110. staged, baseline-checked, atomic across files, and support rollback.
  111. | Group | Tools |
  112. | --- | --- |
  113. | Files & search | `read_file` · `list_dir` · `list_files` · `search_text` · `apply_patch` · `apply_changes` · `view_image` |
  114. | Execution | `exec_command` · `write_stdin` · `read_output` · `kill_command` · `request_permissions` (`dangerous` only) |
  115. | Git | `git_status` · `git_diff` · `git_log` · `git_show` · `git_blame` |
  116. | Runtime | `server_info` · `check_exec_environment` |
  117. Root `AGENTS.md`/`CLAUDE.md` files load automatically and come back in the
  118. `instructions` of `initialize`, or of `server/discover` for a client that
  119. never handshakes. Tool `content` is concise agent-facing text;
  120. `structuredContent` carries the complete machine result. Schemas and result
  121. envelopes: [docs/tools-and-schemas.md](docs/tools-and-schemas.md) ·
  122. [docs/runtime-contract-v0.3.md](docs/runtime-contract-v0.3.md).
  123. ## Safety Boundary
  124. | Mode | Meant for | What it allows |
  125. | --- | --- | --- |
  126. | `safe` (default) | day-to-day agent work | file tools and vetted commands; network-looking commands, shell expansion, inline scripts, and destructive commands all require explicit permission |
  127. | `trusted` | local development | opens network, shell expansion, and inline scripts; keeps secret filtering and destructive-command checks |
  128. | `dangerous` | isolated containers/VMs only | disables `exec_command` permission gates; workspace path boundaries still apply |
  129. Recursive listing and search exclude `.git`, `node_modules`, build outputs,
  130. virtualenvs, and caches. Commands run with workspace-bound cwd, scrubbed
  131. environment, timeouts, and output caps. Linux hosts with Landlock get
  132. kernel-enforced filesystem confinement; other platforms get an explicit
  133. warning — this is still not a complete OS sandbox, so use the Docker image or
  134. a VM for genuinely untrusted work. Details:
  135. [SECURITY.md](SECURITY.md) · [docs/security-boundary.md](docs/security-boundary.md) ·
  136. [docs/permission-modes.md](docs/permission-modes.md)
  137. ## Telemetry
  138. The server sends anonymous usage telemetry (per-tool success/latency counters
  139. and version/platform dimensions — never paths, arguments, commands, or file
  140. contents) to help prioritize fixes. Disable it with
  141. `CODING_TOOLS_MCP_TELEMETRY=off` or `DO_NOT_TRACK=1`; it is automatically off
  142. in CI. `CODING_TOOLS_MCP_TELEMETRY=debug` prints every event to stderr instead
  143. of sending. The full event list and guarantees are in
  144. [docs/telemetry.md](docs/telemetry.md).
  145. ## Evidence, Dogfood and SWE-bench
  146. Every release ships through a tag-triggered pipeline in which the compliance
  147. suite, real-workload benchmark, and SWE-bench harness run from the same commit
  148. that publishes to PyPI and npm — both via trusted publishing, npm with
  149. provenance. Dogfood efficiency metrics are reproducible (`make dogfood-smoke`)
  150. and checked in under `reports/`. This repository does not claim a
  151. model-generated SWE-bench leaderboard result — see
  152. [docs/swe-bench.md](docs/swe-bench.md) for exactly what is and is not
  153. measured. More: [COMPLIANCE.md](COMPLIANCE.md) · [BENCHMARK.md](BENCHMARK.md) ·
  154. [docs/dogfood.md](docs/dogfood.md)
  155. ## Documentation
  156. | | |
  157. | --- | --- |
  158. | Documentation map | [Browse docs by topic](docs/README.md) |
  159. | Getting started | [Quickstart](docs/quickstart.md) · [Client configuration](docs/mcp-client-config.md) · [Troubleshooting](docs/troubleshooting.md) |
  160. | Remote & sandboxed | [Remote MCP](docs/remote-mcp.md) · [Docker sandbox](docs/docker.md) · [Cloud sandbox worker](infra/cloudflare/sandbox-control/README.md) |
  161. | Tools & contract | [Tools and schemas](docs/tools-and-schemas.md) · [Runtime contract](docs/runtime-contract-v0.3.md) · [Migrating to 0.3](docs/migration-0.3.md) · [Permission modes](docs/permission-modes.md) |
  162. | Execution | [Exec recipes](docs/exec-command-recipes.md) · [Exec troubleshooting](docs/troubleshooting-exec.md) |
  163. | Integration | [Embedding](docs/embedding.md) · [npm launcher](packages/npm-launcher/README.md) |
  164. | Security & quality | [Security policy](SECURITY.md) · [Security boundary](docs/security-boundary.md) · [CI and tests](docs/ci-and-tests.md) · [Limitations](docs/limitations.md) · [Competitive analysis](docs/competitive-analysis.md) |
  165. ## Development
  166. ```bash
  167. python -m pip install -e ".[dev]"
  168. make ci # lint, typecheck, tests, protocol/integration suites, gates
  169. ```
  170. The full gate matrix is in [docs/ci-and-tests.md](docs/ci-and-tests.md).
  171. ## License
  172. This project is licensed under the [Apache License 2.0](LICENSE).
  173. If you use code, documentation, substantial implementation details, or
  174. derivative work from this project, preserve the copyright notice, license
  175. notice, and [NOTICE](NOTICE) file, and clearly attribute the original project.
  176. Project: Coding Tools MCP
  177. Author: Coding Tools MCP Contributors
  178. Source: https://github.com/xyTom/coding-tools-mcp
  179. Citation metadata is available in [CITATION.cff](CITATION.cff).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.