Docs

Setup in two minutes

AI Babysitter is a remote MCP server. Connect your agent by URL, sign in once in your browser, and it checks a curated engineering playbook before it builds — so it stops guessing.

1. Connect by URL (recommended)

Subscribe, then add the server to your harness — no API key needed. In Claude Code:

claude mcp add --transport http aibabysitter https://aibabysitter-production.up.railway.app/mcp

Then run /mcp, select aibabysitter, and choose Authenticate — log in with your AI Babysitter email and password and click Approve. Subscribers get per-harness instructions in the setup guide.

2. Headless servers & CI (API key method)

No browser available, or your harness doesn't support remote MCP with OAuth? Generate an API key from your dashboard (shown once — save it; regenerate anytime) and run the stdio package instead. Claude Code: add this to ~/.claude/mcp.json. Any MCP-capable harness (Cursor, Windsurf, custom agents) works the same way.

{
  "mcpServers": {
    "aibabysitter": {
      "command": "npx",
      "args": ["-y", "@aibabysitter/mcp-server"],
      "env": {
        "AIBABYSITTER_API_KEY": "aibab_...your key...",
        "AIBABYSITTER_API_URL": "https://api.aibabysitter.ai"
      }
    }
  }
}

3. Tell your agent to use it

Add this to your project's AGENTS.md / CLAUDE.md so the agent consults the knowledge base when durable project context could affect its approach:

## Knowledge Base Usage
When repository-specific lessons, prior decisions, or durable context could affect a task:
1. Call query_knowledge_base with the relevant question.
2. Use returned lessons to ground your approach.
3. If retrieval fails, report the exact error rather than assuming context was available.

The tools

The default public MCP surface exposes exactly two tools:

query_knowledge_base(query, partition?)

Search curated lessons when repository-specific context, prior decisions, or durable project knowledge could change the approach. Returns up to 5 ranked results with title, content, and tags.

get_usage()

Current plan, monthly call usage, and quota reset date.

Preview write access — not in the default tool list

push_lesson can write a short lesson or decision after a task only when the backend deployment enables knowledge writes and your tenant has the knowledge-write entitlement. For the stdio client, it also requires MCP_KNOWLEDGE_WRITES_ENABLED=true.

Troubleshooting

401 Unauthorized — for the URL method, re-authenticate via /mcp. For the API-key method, the key in AIBABYSITTER_API_KEY is missing or was regenerated — grab the current one from your dashboard.
403 Subscription inactive — your payment failed or the plan was cancelled. Fix billing from the dashboard's "Manage billing" button.
429 Quota exceeded — you hit the monthly fair-use ceiling; it resets on your quota date (see get_usage).

Stuck? Email support@aibabysitter.ai.