@brightbeamai/chap-coordinator-a2a

A2A server adapter for the CHAP Coordinator. Wraps a Coordinator instance and exposes every CHAP method as an A2A AgentSkill on the Agent Card, so any A2A-aware orchestrator (Azure AI Foundry, Amazon Bedrock AgentCore, Google ADK, custom multi-agent systems) can register the coordinator by URL and delegate work to it.

Spec target: A2A 0.3.0 (the version implemented by @a2a-js/sdk). CHAP 0.2.

Install

npm install @brightbeamai/chap-coordinator-a2a

Node 20+ required. Runtime dependencies: @brightbeamai/chap-coordinator, @brightbeamai/chap-coordinator-mcp (for the shared schemas), and @a2a-js/sdk.

Quick start

import express from "express";
import { Coordinator } from "@brightbeamai/chap-coordinator";
import {
  makeChapAgentCard,
  makeChapAgentExecutor,
} from "@brightbeamai/chap-coordinator-a2a";
import {
  DefaultRequestHandler,
  InMemoryTaskStore,
} from "@a2a-js/sdk/server";
import { A2AExpressApp } from "@a2a-js/sdk/server/express";

const coord = new Coordinator({
  defaultProfiles: ["core/1.0", "review/1.0", "audit-scitt/1.0"],
});

const card = makeChapAgentCard({ baseUrl: "http://localhost:9090" });
const executor = makeChapAgentExecutor(coord);

const handler = new DefaultRequestHandler(card, new InMemoryTaskStore(), executor);
const app = new A2AExpressApp(handler).setupRoutes(express(), "");
app.listen(9090);

A runnable reference server is at reference/a2a-server-ts/.

What gets exposed

All 39 CHAP methods become A2A skills named chap.<method>, matching the MCP transport's tool naming so callers fluent in one are fluent in the other.

A2A messages carry the CHAP params in a DataPart. The skill id identifies which CHAP method to dispatch, looked up in this order: message.metadata.skill first, then part.data.skill on the first data part.

Spec version asymmetry

The two official A2A SDKs are at different points of the spec evolution. This adapter targets A2A 0.3.0 via @a2a-js/sdk. The Python counterpart in chap_coordinator.transports.a2a_server targets A2A 1.0 via a2a-sdk. The CHAP adapter layer is identical across both; the spec-version difference is a property of the SDK ecosystem we depend on. Both Agent Cards advertise the correct version for each implementation.

Architecture

• One Coordinator, one A2A agent. Multi-workspace handled inside the Coordinator.
• Stateless adapter. Each message/send translates to a JSON-RPC envelope and dispatches through coord.dispatch().
• Auth deferred. A2A's security schemes attach to the Agent Card and are enforced at the HTTP transport.

Tests

npm test

The integration tests cover the Agent Card shape, dispatch via data part, dispatch via metadata, error surfacing, the full workflow including override, deliberation flow, and cancellation.

See also

• The integration narrative: integrations/CHAP-with-A2A.md
• The walkthrough: examples/drive-chap-from-an-a2a-orchestrator.md
• The Python counterpart: chap_coordinator.transports.a2a_server

License

Apache 2.0. See the parent repository's LICENSE.