Connect your agent

Crilo exposes a remote MCP endpoint per workspace. Any MCP-speaking client can drive your projects through it — and the primary pattern is a worker on a loop or schedule.

Add the endpoint

Copy the snippet from your workspace home (the Connect your agent card) — it already has your workspace id baked in:

claude mcp add --transport http crilo https://mcp.crilo.app/mcp/<workspace-id>

For Cursor, Windsurf, or another client, use the same URL with that client’s “add MCP server” flow.

What the agent can do

Once connected, the agent has a focused, task-first tool set:

  • Orient — get_plan gives the whole workspace at a glance (every project, its purpose and live status, the queue, what’s blocked or stale); search recalls what’s already been decided so the agent extends prior work instead of duplicating it.
  • Plan — create projects and goals, set each goal’s state, and order the queue. Goals can be any altitude — “fix the 404 on /pricing” or “increase trial→paid conversion.”
  • Work the queue — claim_next_task atomically takes the top goal and flips it to doing, stamped with the agent, so two looping agents never double-take the same work. Every run ends with a report — done, blocked (with the question for you), or released back to the queue — then a one-line set_status.

The worker loop

The default for an unattended run: get_plan → claim_next_task → work it in your repo → report → repeat. You write the goals and order the queue; agents claim them top-down on a loop or schedule and report back. You watch every status and state change live in the workspace — nothing is hidden behind a generate-and-export step.