Skip to main content

Module server

lc_a2a

Module server 

Source
Expand description

A2A Server - handler functions for the Agent-to-Agent protocol.

Provides A2AServer which holds an underlying agent (a BaseChain) and exposes handler functions that can be plugged into any HTTP framework (axum, actix, warp, etc.) rather than running its own server.

§Endpoints

  • GET /.well-known/agent-card.json -> returns AgentCard (via get_agent_card)
  • POST / -> accepts A2ARequest, dispatches, returns A2AResponse (via handle_a2a_request / handle_a2a_request_authenticated)

§Task Model

tasks/send follows the A2A asynchronous task lifecycle. The request is acknowledged immediately with a submitted task and the chain runs in the background, transitioning the task submitted -> working -> completed (or failed). Poll tasks/get to observe progress. Every transition is guarded by the TaskStatus state machine, so a task cancelled while the chain is still running is never clobbered back to a live state.

§Multi-turn & Input-Required (P2-2/P2-3)

Re-sending tasks/send with a taskId appends a message to the existing task’s history and re-runs the chain over the whole conversation. A chain that needs more information returns a ChainError::MissingInput / ChainError::InputError, which the server maps to the input-required state; the client then resumes with tasks/send {taskId, message}.

§Ownership & Idempotency (P1-4/P1-6)

Tasks carry an optional owner taken from request metadata. tasks/get and tasks/cancel from a caller whose metadata owner does not match the task’s are rejected (-32003). A message_id in request metadata makes tasks/send idempotent: re-sending the same id returns the already created task instead of running the chain twice.

§Task Persistence (P1-1)

Tasks are stored through the TaskStore trait, defaulting to an in-memory InMemoryTaskStore shared with background workers. Swap in your own backend with A2AServer::with_store. Terminal tasks older than the configured TTL are cleaned up lazily on read access.

§Streaming (P2-1)

Enable A2AServer::with_streaming to get a broadcast channel of TaskPushNotifications (subscribe()), which an HTTP layer can expose as an SSE endpoint. The agent card then advertises {"sse": true}.

§Skill routing (P2-4)

A2AServer::with_skill_router dispatches tasks/send requests that carry a skillId param to a different chain based on the card’s skills.

§Example

ⓘ
use lc_a2a::{A2AServer, AgentCard};
use lc_chains::LLMChain;
use std::sync::Arc;

let chain = Arc::new(LLMChain::new(llm, "You are a helpful assistant"));
let server = A2AServer::new(chain)
    .with_card(AgentCard::new("my-agent", "A helpful agent", "http://localhost:8080"));

// In your HTTP handler:
let response = server.handle_a2a_request(request).await;

Structs§

A2AServer
A2A Server - wraps an agent and provides handler functions.