# agentmods-mcp

An [MCP](https://modelcontextprotocol.io) server for
[agentmods.dev](https://agentmods.dev) — a catalogue of 648,000+ mods for
coding agents (skills, subagents, slash commands, hooks, MCP servers, Cursor
rules, instructions files) where every entry is **measured**: token cost per
session, a security scan with the evidence line for each finding, original
vs copy, and what changed since last week.

Plug it into Claude Desktop, Claude Code, or any MCP client, and the model
can search the catalogue and choose by the measurements — not by vibes.

## Setup

`.mcp.json` (project) or your client's MCP config:

```json
{
  "mcpServers": {
    "agentmods": { "command": "npx", "args": ["-y", "agentmods-mcp"] }
  }
}
```

Zero dependencies, Node >= 18, stdio transport.

## Tools

- **search_mods** `{ query, type?, limit? }` — names ("playwright") and plain
  intent ("make my agent cheaper") both work; every result carries
  `tokens_per_session`, `security_grade`, `origin`, `repo_stars` and the
  install command.
- **mod_info** `{ slug }` — one mod's full detail: measurements, licence,
  files, and per-agent install commands (Claude Code, Codex, Gemini CLI,
  Cursor, GitHub Copilot).

Agents **with** a shell can skip MCP entirely: the
[`agentmods` CLI](https://www.npmjs.com/package/agentmods) searches and
installs directly (`npx agentmods add <slug>`), and this server ships inside
it too (`npx agentmods mcp`).

MIT licence.