Start with the two systems that need to communicate. The protocol name is the second decision.
| Connection job | Start with | What the protocol is responsible for |
|---|---|---|
| Application or frontend to Awaken | Managed Agents, AI SDK, or AG-UI | Turn an application request into input and events for one Awaken Session |
| Agent to tool or data service | MCP | Export a reviewed Awaken tool set or attach an external MCP server to an Agent |
| Agent to Agent | A2A | Discover a remote Agent and exchange task state across an HTTP boundary |
| Worker to an external Agent process | ACP | Let a Worker run a supported external CLI as the Brain for one governed execution |
| Operator to a Session already running | Live Inbox | Change queued input before the Agent consumes it |
| Awaken to your backend | Webhooks | Send signed lifecycle notifications so a backend can fetch current resource state without polling |
These are adapters around one published Agent and Session model. Choosing a different wire does not create a second execution path, permission system, or event history.
Continue with the exact connection
Use the connection matrix for the authoritative direction, endpoint, configuration surface, authentication rule, and observable completion signal. Those details live there so they do not drift across protocol introductions.
After choosing a row, open its protocol guide to understand the boundary, then use the linked how-to guide to make one recognizable request and verify the same Session or event record.