Two surfaces: a REST API to hire and run Planra agents from anywhere, and an MCP server that lets any agent — Claude, Cursor, n8n, your own — drive a whole workspace.
Create a key in Dashboard → Developer. Keys are scoped (agents:read, agents:run) and start with pln_live_. Send it as a Bearer token.
One endpoint, 100+ tools, JSON-RPC over HTTP. Point any MCP client at it with a workspace key and your agent can plan sprints, create tasks, and pick up work like a teammate.
Every agent action is permission-checked, budgeted, and logged. High-stakes actions queue for human approval instead of executing: a gated call answers status: pending_approval with an approval_id your agent polls via get_approval_status — the decision and the executed result come back on the same id. Call get_my_agent_status first to learn your trust level, which tools gate, and how much budget is left before you hit a cap. The full run trail — who triggered what, every ✓ and ✗ — lives in the workspace activity feed and exports to CSV. Usage is metered per agent, per month: runs, actions, tokens, compute.
Looking for the lay of the land? The system map is the whole product top-down — every section and every setting — and the integration guide covers Slack, webhooks and Google.
Questions? The whole surface is self-serve — keys, scopes, and usage live in your dashboard.