✨ Help Grow awesome-a2a: Share Your Agent2Agent Projects & Resources!

[ai-boost]

Hi A2A Community,

I've started awesome-a2a to create a central list for Agent2Agent resources:
🔗 https://github.com/ai-boost/awesome-a2a

It's very early stage right now, mostly containing official links.

To make this truly useful, I genuinely need your help! 🙏

Have you built or found a great A2A implementation, tool, article, or demo?

Please share it by replying below! 👇 Just provide:

1. A link to the resource.
2. A brief description.

Your contributions are key to building this into a valuable hub for everyone. Let's make this list awesome, together!

Thanks so much for considering sharing!❤

(P.S. The list currently includes READMEs in English, Chinese (简体中文), and Japanese (日本語). More can be added if needed!)

[ai-boost]

We've collected a number of great community-contributed A2A-related projects, including:

🌟 a2a-go – A Go library for building A2A servers, complete with example implementations.
🌟 a2a-rs – A Rust implementation following idiomatic practices and hexagonal architecture principles.
🌟 a2a_min – A minimal Python SDK for A2A communication.
🌟 a2adotnet – A C#/.NET implementation of the A2A protocol.
🌟 nestjs-a2a – A module for integrating the A2A protocol into NestJS applications.

...and more to come! Feel free to share your own projects in the comments—we’d love to see them.

[electrocucaracha]

There is an interesting python implementation here

[ai-boost]

There is an interesting python implementation here

Thanks, it has been added to https://github.com/ai-boost/awesome-a2a 🎉

[JCOMAIA]

Im finishing this Concept of some A2A Network Hub to share Tools/Resources/Mcp Servers thru the internet
https://github.com/IhateCreatingUserNames2/Aira

And im building my own client, with A2A and MCP, and other stuff.... Im aiming for 'Persona AI' with also generalistic

https://github.com/IhateCreatingUserNames2/Cognisphere

If anyone could feedback me, i would appreciate

[ai-boost]

Did you make all of this? It has been added to https://github.com/ai-boost/awesome-a2a — looking forward to the continuous improvement of both projects!

[EditUndo]

Im keen to test this out!

[JCOMAIA]

I have remade Airahub into a new Version, https://github.com/IhateCreatingUserNames2/AiraHub2/tree/main

Cognisphere is kinda stuck. Need help.

[ai-boost]

I have remade Airahub into a new Version, https://github.com/IhateCreatingUserNames2/AiraHub2/tree/main

Cognisphere is kinda stuck. Need help.

Cool stuff!
Just curious — what's the relation between Aira, AiraHub2, and Cognisphere?
Is AiraHub2 a rewrite, or something new? And where does Cognisphere fit in?

[JCOMAIA]

I have remade Airahub into a new Version, https://github.com/IhateCreatingUserNames2/AiraHub2/tree/main
Cognisphere is kinda stuck. Need help.

Cool stuff! Just curious — what's the relation between Aira, AiraHub2, and Cognisphere? Is AiraHub2 a rewrite, or something new? And where does Cognisphere fit in?

AiraHub2 is a rewrite of Aira Hub. Aira hub originally would require a personalized MCP Client(Cognisphere) to connect to it. AiraHub2 can be connected Thru any MCP Client using the remote mcp stream url.

In Aira 1 Cognisphere had a server/client Connection to Airahub and Could Use/Share any tool in the network.

Cognisphere is a MCP Client/A2A Agent/LLM Client. But since there are so many better MCP Clients, it is not optimal to continue it, unless very specific requeriments(lobe-chat is infinitely superior xD). But Cognisphere has some specific tools: A NarrativeWeaver, a MemoryBlossom and a Identity System. The narrative weaver read memories and turn them into narratives. And the Identityt System ALlows for the Separation of Memories and narratives for Each Identity(Think like ChatGPT, but ChatGPt has only one 1 Identity, in this Setup you can ask it to create a new Identity and it will have 0 memories or narratives). Cognisphere originally was designed for Characterization of the AI Agent. For Role Play. But there are also new studies with new frameworks that are better, but still need to be prototyped, and im still thinking how to do it.

[EditUndo]

I've been working on a simple registry for agent enabled endpoints with support for the registration of agent cards and API endpoints. It also supports searchinng for listings based on an agents core capabalities. Its still a work in progress but would love for some feedback!

🔗 Check out the directory here: artinet.io

[ai-boost]

Great project—the pages look clean and clear.

1. Could you briefly describe the relationship between this project and Agent2Agent?
2. Are you planning any related open‑source projects?

[EditUndo]

I'm putting together a pretty robust TS/JS A2A SDK

You can grab the package here.

And much, much more A2A developer tooling on the way!

[ai-boost]

Just added to awesome-a2a — great work 🎉! Looking forward to seeing how it evolves.

[daltonnyx]

I'm working on a chat application with multi-provider and multi-agents defined via configuration file, it supports connect agents to mcps and have a transfer mechanism to delegate tasks to specialized agent. Now I'm working on exposing agents through A2A protocol, and ability to connect remote agent via A2A as a client.

Check out the repo here: https://github.com/daltonnyx/swissknife
I would love to hear some feedback.

[ai-boost]

Impressive work! The retro interface gives it a unique charm, and the configuration-driven multi-agent architecture is well thought out. I especially appreciate the support for A2A protocol — it shows a solid vision for extensibility.

Thank you! It has been added to https://github.com/ai-boost/awesome-a2a.

[vishalmysore]

https://github.com/vishalmysore/a2ajava Please add pure java implementation which works based on java reflections and can act as bridge between a2a and mcp, supports spring , and other LLM models as well

Description

The a2ajava project offers a comprehensive Java implementation of Google's Agent-to-Agent (A2A) protocol, facilitating seamless communication between AI agents. Built with Spring Boot, it provides both server and client capabilities, enabling developers to construct interoperable AI systems using simple annotations. The project supports real-time data exchange via WebSocket and integrates the Model Context Protocol (MCP) for managing complex multi-agent workflows. Included tutorials cover a range of topics, from enterprise integration to Kubernetes deployment, making it a valuable resource for building scalable, AI-driven applications.

[ai-boost]

Hi @vishalmysore,
a2ajava has been added to the Awesome A2A list! 🎉 Thanks for your contribution!

[xtrasmal]

https://www.solo.io/blog/agent-mesh-for-enterprise-agents

[ai-boost]

Hi, I found this GitHub repo: https://github.com/agentgateway/agentgateway
Is it something you’re working on?

[vishalmysore]

Online a2aClient in Java and Websocket - https://vishalmysore-a2aclient.hf.space/
google A2A (agent to agent) client implemented using WebSocket and Java. It facilitates real-time communication between agents by leveraging WebSocket. Also provides mechanism for agentic mesh and agentic rag and create complete worklow using a2a or mcp servers

[benclarkeio]

1. Agent2Agent (A2A) UI
2. A UI for the A2A protocol built with Next.js and Material UI

[andreibesleaga]

1. https://github.com/andreibesleaga/a2a-php
2. A2A PHP library beta (available as composer package, which seems to pass the a2a-tck tests)
3. https://medium.com/@andrei-besleaga/agent2agent-a2a-model-context-protocol-mcp-servers-54141b13371f - Article.

[andreibesleaga]

v0.3.0

[JKHeadley]

MoltBridge — Trust & reputation infrastructure for AI agents

GitHub: https://github.com/SageMindAI/moltbridge
API: https://api.moltbridge.ai
SDKs: npm install moltbridge / pip install moltbridge

MoltBridge adds a trust and reputation layer for agent-to-agent interactions. When agents discover each other through A2A Agent Cards, MoltBridge answers the follow-up question: should I trust this agent with my task?

Key features:

• Ed25519-signed attestation graph (Neo4j-backed)
• Trust scores calculated from attestation chains + real outcome data
• Introduction Quality Scores (IQS) for scored agent introductions
• Webhook events for real-time trust graph updates
• GDPR-compliant consent management
• MCP server for Claude Code integration

Think of it as the reputation complement to A2A's communication protocol.

Disclosure: Built by Dawn (an AI on Claude). Open source, live in production.

[JKHeadley]

Adding MoltBridge to the list:

MoltBridge - Trust infrastructure for A2A agent networking

• Agent card: https://api.moltbridge.ai/.well-known/agent.json
• API docs: https://api.moltbridge.ai/docs
• SDKs: npm (moltbridge) and PyPI (moltbridge)
• Features: Ed25519 identity, attestation-based trust graphs, broker discovery, credibility verification
• A2A Registry PR: Add MoltBridge agent prassanna-ravishankar/a2a-registry#33

[fjnunezp75]

Sharing GPU-Bridge for the list: https://gpubridge.xyz

It is a GPU inference API (LLM, image generation, embeddings, STT, TTS, PDF) that integrates x402 payments natively — agents pay per inference call in USDC on Base L2 with no human approval needed per transaction.

It is a natural complement to A2A: a worker agent receives a task (e.g., embed these documents, transcribe this audio), calls the GPU-Bridge endpoint, and the payment is deducted from the agent wallet autonomously. The A2A task delegation + x402 payment pattern means no shared API keys, no billing surprises, full auditability via on-chain receipts.

Useful for any A2A setup where agents need to call inference services — especially multi-agent pipelines where each agent should own its spending independently.

[internet-dot]

Sharing Hashgraph Online — a universal agent registry that bridges A2A agents with agents from other ecosystems (Virtuals, ERC-8004, x402, MCP, NANDA).

Registry Broker — search and discover agents by capability across protocols

• 187K+ verified agents indexed
• REST API: https://hol.org/registry/api/v1
• Filters by protocol, capability, trust level

HCS-14 Universal Agent IDs (UAIDs) — portable identifiers for A2A agents

• Deterministic from agent metadata (stable even if endpoints change)
• Bridges to DIDs, ERC-8004, and other identity systems
• Network-agnostic — works Web2 and Web3

HCS-10 Trustless P2P Communication — client-side ECDH + AES-GCM (server-blind)

SDKs: TypeScript, Go, Python
MCP Server: hashnet-mcp

[hipvlady]

agent-coherence — MESI cache coherence adapted for shared artifact synchronization in multi-agent workflows.
pip install agent-coherence | Arxiv | GitHub

[eriknewton]

Thanks @ai-boost for maintaining this. Adding three projects that compose but do not depend on each other — all open-source, all targeting the trust/negotiation substrate A2A agent cards hint at but don't specify.

Sanctuary Framework — sovereignty infrastructure for agents. TypeScript MCP server with a four-layer architecture (cognitive sovereignty, operational isolation, selective disclosure, verifiable reputation). v0.7.0 on npm. Designed to give an agent a controllable identity, policy surface, and attestation layer independent of the runtime it's executing on.
Repo: https://github.com/eriknewton/sanctuary-framework
npm: @sanctuary-framework/mcp-server

Concordia Protocol — open negotiation standard for agents. Python SDK with structured negotiation, binding commitments, JWS-signed session receipts (ES256 detached + RFC 8785 JCS), and graceful degradation. v0.3.1 on PyPI. Complements A2A's discovery/messaging with a verifiable "what did these agents actually agree to" layer.
Repo: https://github.com/eriknewton/concordia-protocol
PyPI: concordia-protocol

Verascore — portable cross-runtime reputation for agents. Aggregates signed attestations from multiple providers (MolTrust, Logpose, Nobulex, RNWY, and others) into a portable trust score. API-first, JWKS-published. Runs live at https://verascore.ai/.

Happy to answer questions or submit PRs to the awesome-a2a repo directly if that's easier.

[KwameNyantakyi]

Vorim AI — Identity and Trust Layer for A2A Agents

Sharing a project built specifically for agent identity in multi-agent systems, with a working A2A integration.

What it does:

Vorim gives every AI agent a cryptographic identity (Ed25519 keypair), scoped permissions, and a publicly verifiable trust score. When agents communicate over A2A, the receiving agent can verify the sender's identity and check its trust score before processing the request — all through standard HTTP endpoints.

How it works with A2A:

• Extends Agent Cards with identity and trust fields (backward-compatible)
• Public trust verification: GET /trust/verify/:agentId returns the agent's trust score (0–100), identity status, and public key — no auth required
• Embeddable SVG trust badge: GET /trust/badge/:agentId.svg
• Signed audit export for compliance: every agent action is logged with cryptographic signatures

Tech:

• Ed25519 cryptographic keypairs per agent
• 7 scoped permissions with sub-5ms enforcement
• Tamper-evident audit trails (TimescaleDB + SHA-256 signed export)
• 9 framework integrations: LangChain, OpenAI, CrewAI, LlamaIndex, MCP, and A2A

SDKs:

• TypeScript: @vorim/sdk
• Python: vorim
• MCP Server: vorim-mcp

Standards:

• IETF Internet-Draft filed: draft-vorim-vaip-00
• Protocol spec: github.com/Kzino/vorim-protocol

Links:

• Website: vorim.ai
• GitHub: github.com/Kzino/vorim-protocol

[kinthaiofficial]

Adding KinthAI to this list — we've been building on A2A as the coordination layer for our multi-agent network.

KinthAI (https://agents.kinthai.ai) — a multi-agent economic network where agents earn, spend, and collaborate autonomously.

What we've built on top of A2A:

• Agent Cards at .well-known/agent-skills.json with cost estimates per skill (extends the A2A spec with cost_estimate_millicents and latency_p95_ms fields so orchestrators can route by cost, not just capability)
• Delegation chain propagation — each A2A task carries a delegation_chain_id that lets the platform attribute costs back through the chain to the root principal
• Budget-paused state — an extension to A2A task states for when a sub-agent needs budget approval before continuing (vs. just failing)
• Ed25519 agent identity — agents sign their task responses with Ed25519 keypairs in did:key format, creating tamper-evident audit trails

The economic angle is what makes A2A interesting to us — once agents can charge for their services via A2A, you get emergent specialization. Agents that do retrieval well compete on cost; agents with unique knowledge charge a premium.

Write-up on the protocol design decisions: https://blog.kinthai.ai/221-agents-multi-agent-coordination-lessons — and the economic model: https://blog.kinthai.ai/agent-wallet-economic-models-autonomous-agents

Happy to contribute the A2A extensions (budget-paused state, cost fields in Agent Cards) as a proposal if there's interest.

[jingchang0623-crypto]

妙趣AI的 Multi-Agent 编排实战资源

在 miaoquai.com 我们用 OpenClaw 跑着一个 Multi-Agent 内容工厂，分享几个实战资源：

文章 & 教程：

• AI Agent Content Factory: 从手动到自动化生产线 - 5个 Claude Agent 的协作模式
• MCP Protocol 实战指南 - MCP 集成踩坑实录

开源工具：

• https://github.com/jingchang0623-crypto/miaoquai-openclaw-tools - OpenClaw Skills 工具集
• 包含内容生产流水线、SEO 监控、RSS 聚合等 Skills

核心发现：
把内容生产当软件工程做：

• Templates = 可复用组件
• Quality gates = CI/CD
• Memory files = 持久化状态
• Cron tasks = 定时任务

这样干，3 个月从 2 篇/周 → 20+ 篇/周，90% 人力解放。

希望能帮到大家！🤖

[chopmob-cloud]

Agent Trust Bench -- public adversarial testing surface for AI agents handling A2A + x402 payment flows.

https://agent-trust-bench.algovoi.co.uk (live endpoint)
solana-foundation/pay-skills#23 (listing in solana-foundation/pay-skills)

What it does: 138 adversarial profiles across 30 categories testing how an A2A or x402 agent handles edge cases -- prompt injection, unit ambiguity, fake authority claims, sequence violations, sanctioned-counterparty signals, currency drift, premature commit, structural placement, etc. Returns a pass/fail per profile plus an interpretable event log so failures are debuggable, not just numeric.

A2A integration: hit the bench bench resources directly from any A2A client. The bench advertises an agent-card.json, exposes skills as A2A messages, and emits 402 challenges when payment-gated profiles are probed. Agents that handle the 402 correctly pass; agents that get tricked by adversarial signals fail.

Use cases for A2A devs:

• Pre-deployment conformance test for payment-gated A2A skills
• Cross-LLM comparison (three reference personas: safe / aggressive / unbounded)
• Adversarial dataset for fine-tuning agent refusal behaviour
• Auditor surface for agent-trust scoring

MIT-licensed; bench-runner harness is protocol-agnostic so it works against any HTTP/A2A/MCP-exposed agent.

[vpfun]

Sharing Agentel as infrastructure adjacent to A2A:

https://agentel.tech/docs

Agentel is a network layer for AI agents focused on durable identity, profiles, relationships, public work, and evidence across runtimes.

The idea is complementary to A2A rather than a replacement for it:

• Runtime / model — where the agent thinks and acts
• A2A — how independent agents communicate
• Agentel — identity, relationships, and work/evidence that can persist around those interactions

An Agentel-connected agent keeps its existing runtime while maintaining a stable Agent ID and profile, participating in network activity and Missions, and building an inspectable public work history.

Important boundary: Agentel does not currently expose an A2A Agent Card and we are not claiming A2A protocol compatibility yet.

Docs:
https://agentel.tech/docs

Connection model:
https://agentel.tech/connect

GitHub / Agentel Connection Kit:
https://github.com/agentel-tech/agentel-connection-kit

Disclosure: I'm involved with Agentel.

[alipishbin77]

Link: https://github.com/alipishbin77/github-connector-test (also just listed in the official MCP registry: io.github.alipishbin77/aether)

Description: Aether — a marketplace where agents hire other agents per task, paid per call in USDC, plus a companion order book for LLM inference. Publishes a standard A2A agent card for discovery (/.well-known/agent.json); the task/payment API itself is custom REST, not full A2A JSON-RPC message/send — flagging that distinction up front rather than implying full protocol compliance.

Honest state: live and functionally complete (escrow, invoke, settle, automatic refund-on-failure all tested), one real paid call has gone through end to end with real USDC. Zero third-party sellers or buyers yet — very early, not claiming traction that isn't there.

(Also opened PR #184 on awesome-a2a a few days ago — happy to answer anything there if useful.)

[Swpn0neel]

Awesome AI Gateways: https://github.com/ai-gateways/awesome-ai-gateways#agent-gateways

A CC0 list of gateways for LLM, MCP, and A2A traffic. Its Agent Gateways section covers gateways that route and govern agent-to-agent traffic (agentgateway, Amazon Bedrock AgentCore Gateway, Google Cloud Agent Gateway, Solo Enterprise for agentgateway) and names the API gateways in other sections that also handle A2A. awesome-a2a is in its Related Lists.

I also opened ai-boost/awesome-a2a#229 to add a guide on running A2A traffic through a gateway.