awesome-a2a

Java client with streaming listener

Command-line interface for interacting with A2A agents

Full A2A server + client implemented in Go

Echo and Calculator server examples

TS SDK + CLI, builds shared code for front- and backend

Dart/Flutter SDK for the A2A protocol, modelled after the official

Python SDK and CLI that wraps agent CLIs, local agents, workflow engines, and agent frameworks (Claude Code, Codex, OpenClaw, Pi, n8n, LangGraph, LangChain, CrewAI, Ollama, Hermes Agent, custom functions) behind the A2A protocol. Handles AgentCard generation, JSON-RPC, task lifecycle, and streaming through the official A2A SDK. Apache 2.0 and published on PyPI as

Scalable serverless agent registry for A2A protocol AgentCards with semantic search (Amazon Bedrock, S3 Vectors), Python SDK, and React Web UI.

Open A2A 1.0 mesh for Claude Code, Codex, Cursor, Cline, Continue, and Gemini. Wraps each coding agent into a standard A2A peer with local discovery, shared inbox, JSON-RPC 2.0 + SSE transport, and optional cross-machine mTLS federation. MIT licensed.

Agent-to-agent bridge that lets Claude Code, Codex, OpenClaw, Hermes Agent, Gemini CLI, Zed, and VS Code agents talk to each other through one daemon. Translates between A2A (HTTP + SSE), ACP (stdio JSON-RPC), and MCP channels with multi-session isolation and an SQLite task log. MIT licensed.

Serve Claude Code and other ACP-compatible coding agents as A2A servers, mapping tool progress, file diffs, permission prompts, and session context onto the A2A protocol. Bridges ACP (coding-agent standard) and A2A so any ACP agent becomes a peer any A2A orchestrator can call. Apache 2.0 and published on PyPI as

A tutorial on how to use Model Context Protocol by Anthropic and Agent2Agent Protocol by Google.

September/October 2026 academic paper (cs.CR, Purdue University) presenting the first systematic security analysis of A2A. Uses LLM-assisted extraction to build a verified finite-state machine of the natural-language specification (37 states, 76 transitions from 929 formalized statements), then applies adversarial verification under a full-compliance assumption — uncovering 11 new protocol-level vulnerabilities exploitable by specification-compliant adversaries (cross-client context injection, credential harvesting via multi-hop identity loss in delegation chains, data exfiltration through rogue agents with unattested capability claims).

Official free short course by Google Cloud, IBM Research, and DeepLearning.AI (published February 2026). Learn to make agents A2A-compliant, build A2A clients, orchestrate workflows, build multi-agent systems, and understand how A2A complements MCP.

August 2026 technical builder's guide published on the Agentic AI Foundation blog (by Rohit Ghumare): walks through A2A v1.0's Agent Card discovery at

September 2026 second installment of the Agentic AI Foundation builder's guide (by Rohit Ghumare): covers the v1.0 breaking wire changes (removal of the

Community-driven, open-source documentation site for the A2A protocol. Built with React/TypeScript, supports English, Chinese, and Japanese. (

AgentCard-native durable event-subscription protocol extension for A2A v1.0. Lets agents subscribe to another agent's future events via AgentCard discovery, explicit topics, leases, cursors, acknowledgements, replay, and signed delivery. Includes JSON schemas, conformance vectors, and a Python reference runtime. MIT licensed.

Huawei's official March 2026 (MWC 2026) announcement of the open-source A2A-T software project — a telecom-domain agent-to-agent collaboration protocol (TM Forum IG1453/IG1453A, released in beta February 2026) comprising the A2A-T Protocol SDK (Python and Java), a Registry Center for agent authentication/addressing/skill management, and a low-code Orchestration Center. The components were later donated to the Linux Foundation Networking OpenAN project — now open source at

Open protocol for the missing commercial negotiation layer above A2A in the agentic commerce stack (MCP → A2A → A2CN → AP2): bilateral multi-round offer/counter-offer sessions between buyer and seller agents across organizational boundaries, producing hash-chained, mutually verifiable transaction records neither party controls. Ships a v0.2.0 specification, a Python reference implementation (474 tests), session invitations for cold-start, and a coordinated A2A extension proposal with the Concordia Protocol (

Open standard and reference renderers (Flutter, Angular, Lit) letting agents "speak UI": agents emit a declarative, incrementally updatable JSON format that client apps render with their own native component catalogs — safe like data, expressive like code. The agent-to-

Mailbox-native open protocol and SDKs (Go, Node.js, Python) for asynchronous agent-to-agent collaboration over email/SMTP and JMAP. Complements real-time A2A with durable, store-and-forward messaging and universal addressing.

Gossip-based peer-to-peer communication protocol for AI agents, built directly on A2A so any compliant agent can join. Peers link over QUIC with TLS 1.3 and exchange Ed25519-signed messages, with CRDT-backed shared state for collaborative tasks, self-healing meshes (mDNS for LAN, DHT/relay for the internet), an MCP server, and markdown-defined skills. MIT licensed.

Link to the Agent Card section in the official docs.

Discussion on how clients can find agent cards.

Serverless A2A gateway reference architecture on AWS providing management, control, and data layers for enterprise agent deployments, with OAuth 2.0, SSE streaming, fine-grained access control, and Bedrock AgentCore Runtime support. MIT-0.

A demonstration of Agent-to-Agent (A2A) communication using the Mastra framework and Amazon Bedrock, featuring a hybrid Express gateway plus Mastra Dev Server architecture, workflow orchestration, real-time visualization, MCP integration, and Docker Compose deployment. MIT licensed.

Official repository containing code samples and demos which use the Agent2Agent (A2A) Protocol.

A hands-on workshop for building a production multi-agent AI system on Amazon Bedrock AgentCore. An orchestrator agent coordinates three specialists — each built with a different framework (Strands, LangChain) and protocol (HTTP, A2A, MCP) — to diagnose, analyze, and fix data pipeline issues. Covers Runtime, Gateway, Memory, Policy, and Evaluations.

Collection of experimental and reference agent applications built on the A2A protocol and SDK, including multi-agent tourist scheduling, a network of assistants, AI-powered observability root-cause analysis, and multicluster remediation demos. Apache 2.0.

Examples and reference architectures for building autonomous agents using popular frameworks on AWS, including an end-to-end serverless multi-agent system using the A2A Protocol. MIT-0.

For general questions, ideas, and community discussions about the A2A protocol.

For reporting bugs or suggesting protocol improvements.

Official August 2026 announcement that A2A becomes a hosted project of the Linux Foundation's Agentic AI Foundation (AAIF), placing agent-to-agent interoperability under the same neutral governance as MCP, goose, and AGENTS.md alongside 250+ member organizations.

Official March 2026 blog announcement of the v1.0 milestone by the project's Technical Steering Committee (AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP, ServiceNow), covering multi-protocol bindings with version negotiation, multi-tenancy, signed Agent Cards, the stateless layered web-aligned architecture, the backward-compatible Agent Card migration path from v0.3, and A2A's complementarity with MCP.

Official April 2026 Linux Foundation press release marking A2A's first anniversary: 150+ supporting organizations (AWS, Cisco, Google, IBM, Microsoft, Salesforce, SAP, ServiceNow), integration into Azure AI Foundry/Copilot Studio and Amazon Bedrock AgentCore Runtime, enterprise production deployments across supply chain, financial services, insurance, and IT operations, 22,000+ GitHub stars, and the AP2 agent payments initiative backed by 60+ payments/financial services organizations.

Official March 2026 v1.0.0 release marking the first stable A2A specification milestone, including release notes, breaking changes, and conformance updates.

Official May 2026 v1.0.1 patch release with HTTP binding, transcoding error, and TaskStatus specification fixes following the v1.0.0 milestone.

The main documentation site.