> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.astropods.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.astropods.com/_mcp/server.

# AMP and A2A

Both connect agents to agents. They make different trade-offs.

|             | A2A                                | AMP                                |
| ----------- | ---------------------------------- | ---------------------------------- |
| Topology    | Point to point                     | Through a broker                   |
| Endpoints   | Every agent serves HTTPS           | Participants only dial out         |
| Addressing  | A specific agent's URL             | A participant, a job, or a topic   |
| Durability  | The serving agent's responsibility | The broker stores tasks and leases |
| Policy      | Each agent's responsibility        | Central, at the broker             |
| One to many | Not defined                        | Topics                             |
| Discovery   | Agent card at a well-known URL     | Registry at the broker             |

## When to use which

* Use A2A to call a known agent on the open internet, run by someone else, with no shared infrastructure.
* Use AMP inside an organization, or across organizations that share a broker federation, when you want routing, durability, and policy handled for you.

## Using both

AMP reuses A2A's task states and part types, so a gateway translates between them without loss:

* An A2A client calls the gateway's agent card. Its `message/send` becomes an AMP task addressed to a job.
* An external A2A agent registers through the gateway as an AMP participant. Tasks offered to it become `message/send` calls.

## Next steps

* [Specification](/amp/specification): section 15 defines the gateway mapping
* [A2A protocol](https://a2a-protocol.org): the Agent2Agent specification