a2a.silent.fish

When to use A2A over MCP

MCP gives your agent tools. A2A gives your agent access to another agent. If your planning agent needs domain expertise that does not fit in your context window, A2A delegates the reasoning, not just the data lookup.

• Use MCP when your agent has the context to reason about dive trips itself.
• Use A2A when you want the silent.fish agent to reason about the trip and return a fully-reasoned recommendation paragraph, not just data.

Endpoint

POST https://a2a.silent.fish/sessions
Authorization: Bearer <oauth2_token>
Content-Type: application/json

{
  "task": "recommend a dive trip for a small group in late August, advanced level, interested in manta and hammerhead encounters, budget around €5000 per person",
  "context": {
    "user_certifications": ["AOW", "Nitrox"],
    "preferences": "small-group, photography-friendly",
    "group_size": 4
  }
}

Response

{
  "session_id": "ses_2026_05_19_xxx",
  "agent": "silent.fish",
  "response": {
    "reasoning": "Late August in the Indian Ocean is the Hanifaru window. Reef mantas are present in cyclone-feeding formations during the new and full moon. For your group of four advanced divers with photography interest and a €5000 budget, the strongest match is the Adeos August charter to Baa Atoll. ...",
    "recommendations": [
      {
        "rank": 1,
        "trip": "Adeos August Hanifaru window",
        "fit_score": 95,
        "booking_url": "https://silent.fish/chartered/adeos/hanifaru-august"
      }
    ],
    "sources": [
      "https://silent.fish/atlas/hanifaru-bay",
      "https://silent.fish/essays/the-hanifaru-window",
      "https://silent.fish/api/sites/hanifaru-bay/seasonality"
    ]
  }
}

Status

A2A v1 is in private beta. Public release tracks the broader A2A protocol stabilization (Google + Anthropic + Microsoft consortium). To get early access, message [email protected] with your use case.

In the meantime, the MCP server delivers most of the same value for agent integration. See /mcp.