agent-ops-mcp

A demo MCP server showing the shape of a setup I built for my team: an agent that reads and writes to our internal systems directly, instead of someone pasting context into a chat by hand, plus a lightweight quality gate that checks what the agent produced before it reaches a human reviewer.

The real version talks to our actual task tracker, internal docs, and CI — that code isn't public. This repo is a self-contained rebuild of the same idea against mock data, so it runs standalone with no credentials or access to anything internal.

What it exposes

review_check is the interesting one: instead of a human catching a stray TODO or a hardcoded password during review, the agent runs this on its own output first. It mirrors the review-checklist doc bundled in this repo.

Related MCP server: mcp-docs-tools

Run it

npm install
npm run build
npm start

This starts an MCP server over stdio. To try it from Claude Desktop or Claude Code, add it to your MCP config:

{
  "mcpServers": {
    "agent-ops-mcp": {
      "command": "node",
      "args": ["/absolute/path/to/agent-ops-mcp/dist/index.js"]
    }
  }
}

For local development without a build step:

npm run dev

Data

Everything lives in memory, seeded from src/data/tasks.ts and src/data/docs.ts on startup. create_task mutates the in-memory list — nothing is written to disk, so state resets every run.

Stack

TypeScript, Node.js, @modelcontextprotocol/sdk, Zod for input validation.

License

MIT