MCP Servers

模型上下文协议服务器、框架、SDK 和模板的综合目录。

MCP server exposing developer tools (file I/O, bash, search, web fetch) to LLM agents with security guardrails

创建于 4/10/2026
更新于 about 9 hours ago
Repository documentation and setup instructions

mcp-devtools

An MCP (Model Context Protocol) server that exposes developer tools — file I/O, search, shell execution, task tracking, and web fetching — to LLM agents over HTTP.

Overview

mcp-devtools wraps common developer operations as MCP tools, letting LLM agents (Claude, etc.) perform real work in a controlled environment. It runs as a FastAPI/FastMCP HTTP server and ships with security guardrails to prevent destructive or sensitive operations.

Tools exposed

| Tool | Description | |------|-------------| | read | Read file contents | | write | Write files | | edit | Targeted string replacement in files | | glob | File pattern matching | | grep | Regex search across files | | bash | Execute shell commands (guardrailed) | | web_fetch | Fetch URLs (SSRF-protected) | | todo_write | Manage task lists |

Security guardrails

The server enforces guardrails at the tool layer — no configuration required:

  • Path protection — blocks writes to system paths (/etc, /bin, /System, etc.)
  • Sensitive file blocking — refuses access to .env, SSH keys, TLS certs, cloud credentials, etc.
  • Bash command blocking — blocks destructive commands (rm -rf /, fork bombs, sudo, disk overwrites, curl-pipe-to-shell, etc.)
  • SSRF protectionweb_fetch blocks requests to private IP ranges and cloud metadata endpoints

Installation

Requires Python 3.13+ and uv.

git clone https://github.com/mredvard/mcp-devtools.git
cd mcp-devtools
uv sync

Running locally

uv run uvicorn devtools.app:app --host 0.0.0.0 --port 8000

The MCP endpoint is available at http://localhost:8000/llm (streamable HTTP transport).

Health check: GET /health
Tool discovery: GET /llm/meta/tools

Running with Docker

docker build -t mcp-devtools .
docker run -p 8000:8000 mcp-devtools

Kubernetes deployment

Manifests for a sandboxed Kubernetes deployment are in k8s/llm-sandbox/. See k8s/llm-sandbox/README.md for deployment instructions.

cd k8s/llm-sandbox
bash deploy.sh

Development

uv sync --group dev
uv run pytest

Project structure

src/devtools/
├── app.py          # FastAPI app with MCP mount
├── server.py       # FastMCP instance and tool registration
├── guardrails.py   # Security validation (paths, bash, SSRF)
└── tools/          # Individual tool implementations
k8s/llm-sandbox/    # Kubernetes manifests

License

MIT

快速设置
此服务器的安装指南

安装包 (如果需要)

uvx mcp-devtools

Cursor 配置 (mcp.json)

{ "mcpServers": { "mredvard-mcp-devtools": { "command": "uvx", "args": [ "mcp-devtools" ] } } }