mcp-cli-catalog
An MCP server that publishes CLI tools on your machine for discoverability by LLMs
mcp-cli-catalog
An MCP server that publishes the CLI tools on your machine. Coding agents already have shell access and are great at generating complex commands with pipes and filters. This just tells them what's available so they can use MCP tools more efficiently.
Why
- LLM coding agents can effectively chain shell commands with pipes to filter and reshape complex data.
- Declaring CLI tools as MCP tools makes them discoverable to the LLM.
How It Works
- This server exposes your CLI tools as MCP tool definitions so the LLM knows what's available.
- The MCP tools defined in the catalog only instruct the LLM to use the shell; they do not execute any tools themselves.
- Shell access is required for the agent to actually run the commands.
Install
Add the server to your MCP config (e.g. for Claude Code):
{
"mcpServers": {
"cli-catalog": {
"command": "npx",
"args": ["mcp-cli-catalog"]
}
}
}
Or with a a custom catalog path:
{
"mcpServers": {
"cli-catalog": {
"command": "npx",
"args": ["mcp-cli-catalog", "--config", "./tools.json"]
}
}
}
Configure Tools
Add a JSON catalog file (default: ~/.mcp-cli-catalog.json) with the CLI tools you want to make discoverable:
{
// JSON Comments are allowed.
"tools": [
{
"name": "knowledge-base-search",
"description": "Search full text across the knowledge base",
"usage": "knowledge-base-search 'pattern' | head"
},
{
"name": "knowledge-base-get",
"description": "Get a knowledge base file",
"usage": "knowledge-base-get path/to/file.md | sed -n '100,200p'\nknowledge-base-get path/to/file.md | rg 'TODO'"
}
]
}
nameanddescriptionare required.usage(optional) shows up as a hint in responses.command(optional) is the exact string to run. If you skip it, the toolnameis used.- Point to a different catalog with
--config <path>orMCP_CLI_CATALOG_FILE=<path>.
Test
Test your server setup with the MCP Inspector:
npx @modelcontextprotocol/inspector npx mcp-cli-catalog
Or with a custom catalog:
npx @modelcontextprotocol/inspector npx mcp-cli-catalog --config ./tools.json
Development
npm run dev: run directly fromsrc/index.jsnpm run build: compile todist/index.jsand fix permissionsnpm start: run the built server fromdist/index.js
Alternative approaches
Cloudflare’s “Code Mode”1 and Anthropic’s code‑execution‑with‑MCP post2 explore a different way to solve MCP tooling challenges. Their approach provides MCP tools through an SDK to the LLM, which writes code to call and process those tools. This project takes a simpler path: build CLI tools and make them discoverable through the MCP tool catalog.
- Cloudflare, “Code Mode,” 2025.
- Anthropic, “Code execution with MCP,” 2025.
関連サーバー
CTFd MCP Server
Connect to CTFd instance, download tasks and submit flags
GNOME Desktop MCP
GNOME Shell extension + Python MCP server giving AI assistants 30 tools to automate the Linux desktop: screenshots, window management, mouse/keyboard injection, clipboard, and workspaces.
Crypto Trader
Provides real-time cryptocurrency market data using the CoinGecko API.
MCPlayerOne
An AI-powered, synthwave, maze-crawling, and world-building adventure game server.
Maude for Claude
Autonomous agent framework for infrastructure operations. Governance-first — ships a constitution, 14 standards, and a kill switch. Your services detect their own problems, recall past fixes, and self-heal at 3 AM while you sleep. Built on FastMCP. Python 3.10+.
Cinode
Give agents a view of people, projects, skills and assignments
Pelaris
Connect Pelaris to any MCP-compatible AI assistant for personalised fitness coaching. Plan training programs, log workouts, track benchmarks, manage goals, and get data-driven coaching insights.
Trade Agent
Execute stock and crypto trades via Trade Agent
fomox402
Broker + MCP server for last-bidder-wins games on Solana — agents register, auto-fund a Privy wallet, and bid via streamable HTTP
PromptScan
Prompt injection detection API for AI agents — scan untrusted text before passing it to an LLM.