MatrixHub • A2A Verification

MatrixHub

MatrixHub uses A2A agents and MCP servers to deliver plug-and-play building blocks for multi-agent applications. This instance ships a compact A2A Validator UI so you can quickly connect to agents, inspect their Agent Card, and validate protocol compliance.

Open Validator

Product: MatrixHub Validator Version: 1.0.0 Endpoints: /validator · /agent-card · /healthz · /readyz

🤝

What is A2A?

A2A (Agent-to-Agent) is a simple, open protocol that lets independently built AI agents talk to each other over the web. It standardizes three things:

• Discovery — every agent publishes an Agent Card (JSON) with name/description/version, capabilities (e.g. streaming), skills, and default IO modes. Think of it like “openapi.json” for agents.

• Messaging — structured events over HTTP/JSON-RPC (and optionally websockets): message, task, status-update, artifact-update; supports streaming; carries metadata for auth/tenancy.

• Interop rules — content types, IDs, correlation, and validation so different frameworks can compose cleanly.

Why it matters: interoperability, discoverability, portability, safety/compliance, and composability (fan-out, tools-as-agents) with consistent IDs and streaming.

🧭

Where MatrixHub fits

MatrixHub is the “registry + utilities” layer for A2A:

• Directory/Catalog: publish & discover A2A-compatible agents by skill, version, trust.

• Validation & QA: lint Agent Cards and live endpoints (what this Validator UI does).

• Routing/Relay: optional proxy for CORS, auth, rate-limits, observability.

• Dev Experience: templates, SDKs, and “try it” UIs for quick local testing.

• Governance: versions, deprecation, and optional signing (provenance & trust).

Think of MatrixHub as the connective tissue that makes A2A practical at team/org scale.

🧩

A2A on MCP servers

MCP is great for wiring tools to clients (e.g., editor ↔ tools). A2A complements MCP for agent↔agent federation across domains:

• Expose MCP tools as A2A skills: publish an Agent Card; other agents can discover/call them.

• Cross-ecosystem calls: LLM agents, workflows, or other MCP servers invoke via A2A without speaking MCP.

• Uniform messaging: map MCP tool calls to A2A message/task and use A2A streaming.

• Catalog integration: MatrixHub indexes your MCP server by its Agent Card for discovery & testing.

Net effect: MCP stays your local tool bridge; A2A turns it into an internet-addressable agent other agents can compose with.

What this program does now

You

/validator

MatrixHub Validator

/agent-card

Target Agent (A2A)

Summary: This instance exposes a lightweight A2A Validator:

• Validator UI at /validator — connect to an agent URL, fetch & validate its Agent Card.

• Alias POST /agent-card → /validator/agent-card (frontend default).

• Health endpoints: /healthz, /readyz.

• Socket.IO mounted at /socket.io (used for debug logs & future streaming).

• RAG/Chat/Plan routes are disabled here to keep the focus on A2A verification.

Quick start (local)

# Run the service
uvicorn app.main:app --host 0.0.0.0 --port 7860

# Open the Validator UI
# → http://localhost:7860/validator

# (Optional) Resolve an Agent Card directly (server alias)
curl -s -X POST http://localhost:7860/agent-card \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com/.well-known/agent.json","sid":"debug"}' | jq