Agent access
Give each AI agent its own API key, limited to the clients, services and actions it needs. Every ready-made action is also an MCP tool, so Claude, Cursor, OpenAI's models and agent frameworks can call it directly. The agent only ever holds the key, never the client's token.
- One key per agent or workflow, each scoped and revocable
- An MCP server on the same API, plus a direct HTTPS API
- Every call logged, rate limited, and shown to the client
- ▸list_clients()✓ 1 client
- ▸sheets.read_rows("Bookings")✓ 48 rows
- ▸calendar.create_event(…)✓ created
- ▸slack.send_message(#front-desk)✓ sent
- ▸DELETE hubspot /crm/v3/objects/contacts/…✕ outside this key's limits