# Which AI Agent Frameworks Support A2A? 12, Verified

> Source: The Agents Index — https://theagentsindex.com/blog/a2a-protocol-support-frameworks-2026 (sourced research)
> By The Agents Index Editorial · Published 2026-08-08

_A2A lets agents on different frameworks talk to each other. We checked all 12 frameworks in our index: 4 ship it natively, 3 as an add-on, 4 don't support it yet._

We already answered which agents connect to tools and data through [MCP](/blog/mcp-support-across-ai-agents-2026/). This is the other half of the interoperability question: can an agent built on one framework hand off work to an agent built on a different one, by a different vendor, without custom glue code for that specific pair? That's what the Agent2Agent (A2A) protocol is for, and vendor marketing pages tend to answer it with a single reassuring bullet point. We checked the primary source for all 12 frameworks in our index instead.

## A2A vs. MCP: two different problems

MCP standardizes how an agent reaches a tool, database, or API: one client implementation, any MCP server. A2A standardizes something adjacent but distinct: how one autonomous agent discovers what another agent can do and delegates a task to it, over a shared HTTP/JSON-RPC interface, without either side needing to know the other's internal framework or vendor. Google originated the protocol and later donated it to Linux Foundation governance, and our own [Google ADK listing](/google-adk) documents over 150 organizations reporting production use as of this research.

The distinction matters for a real buying decision. MCP support tells you an agent can be scripted against tools. A2A support tells you an agent can be one node in a system built from agents you didn't write, possibly on frameworks you don't use.

## What we checked, and how

For each of the 12 frameworks in our [frameworks category](/category/frameworks), we went to the vendor's own current documentation site or official GitHub org, not a third-party writeup, and looked for an explicit, dated statement of A2A support. Where a framework's own docs were silent, we checked its GitHub issue tracker for an open or closed feature request, since that's usually the clearest signal of "not yet" versus "quietly shipped." Every row below is sourced to a primary page checked on 2026-08-09.

### Native: built into the framework itself

| Framework | What we found |
|---|---|
| [Google ADK](/google-adk) | A2A is a first-class primitive across the Python, Go and Java SDKs, with dedicated quickstarts for exposing and consuming A2A agents. |
| [Microsoft Agent Framework](/microsoft-agent-framework) | Official packages ship for Python (`agent_framework.a2a`), C# (`Microsoft.Agents.AI.Hosting.A2A`) and Go, for both hosting and consuming A2A agents. |
| [Mastra](/mastra) | Built-in support for A2A protocol versions 0.3 and 1.0 on the same agent card and execution URLs, no external package required. |
| [Agno](/agno) | The AgentOS runtime exposes any Agno agent or team over A2A with a single `a2a_interface=True` flag, added in v2.1.2. |

### Official, but a separate install or hosted product

| Framework | What we found |
|---|---|
| [CrewAI](/crewai) | A2A is an official, opt-in extra (`pip install 'crewai[a2a]'`), described in CrewAI's own docs as "a first-class delegation primitive" once installed. |
| [Pydantic AI](/pydantic-ai) | A2A ships through FastA2A, an official companion library that the Pydantic team has since spun out into its own repository under the `pydantic` GitHub org, rather than bundling it in the core package. |
| [LangChain](/langchain) | The A2A endpoint lives in LangSmith's Agent Server, LangChain's paid, hosted deployment product, exposed at `/a2a/{assistant_id}`. It is not part of the open-source LangGraph orchestration library itself. |

### No official support found

| Framework | What we found |
|---|---|
| [Microsoft AutoGen](/microsoft-autogen) | A maintainer's reply to a GitHub discussion asking about A2A support: "If you would like to help create one, happy to review the contribution." No built-in support, and this is a genuinely different Microsoft product from Agent Framework above. |
| [OpenAI Agents SDK](/openai-agents-sdk) | Two separate GitHub feature-request issues asking for A2A support, neither confirmed as implemented by a maintainer. |
| [Letta](/letta) | No mention of A2A, or of agent-to-agent interoperability generally, anywhere in its current documentation. |
| [Claude Agent SDK](/claude-agent-sdk) | Anthropic's own capabilities list (tools, hooks, subagents, MCP, permissions, sessions, skills, plugins) does not include A2A. Only unofficial, community-maintained wrapper repositories bridge it. |

### Unconfirmed

[LlamaIndex](/llamaindex) posted about building "A2A-compatible agents" on its own account in April 2025, and a sample exists in the protocol's own sample-code repository, but we found no current, maintained documentation page or package on LlamaIndex's own site confirming ongoing official support as of this research. We're counting that as unconfirmed rather than folding it into either "yes" or "no": a year-old social post is not the same claim as a current docs page, and we'd rather say so than round it up.

## The nuance a vendor page won't volunteer

Two things surprised us enough to call out on their own.

First, Microsoft ships two separate multi-agent frameworks in our index, and they land on opposite sides of this question. AutoGen, the older of the two, has no official A2A support and its own maintainer has only offered to review a community contribution. Agent Framework, the newer, unifying SDK, ships A2A natively across three languages. If you've been tracking "Microsoft's A2A story" as one thing, it's actually two, and picking the wrong one changes your answer.

Second, LangChain's A2A support is real, but it sits behind LangSmith's paid Agent Server, not inside the open-source LangGraph library most people mean when they say "LangChain." A team building a fully self-hosted, OSS-only multi-agent system won't get A2A from LangGraph alone the way they would from Mastra or Agno, where the protocol ships in the free, self-hostable core.

## What this means if you're choosing a framework today

If cross-vendor agent interoperability is a real requirement, not a hypothetical one, four frameworks give you A2A without an extra install or a paid tier: Google ADK, Microsoft Agent Framework, Mastra and Agno. CrewAI and Pydantic AI get you there with one additional official package. LangChain gets you there only if you're already paying for LangSmith's hosted deployment. AutoGen, the OpenAI Agents SDK, Letta and the Claude Agent SDK don't have an official answer yet, which doesn't rule any of them out for other reasons, but does mean you'd be building the interop layer yourself or waiting on the vendor.

## Methodology & sources

Every claim above is sourced to a primary page: the framework's own current documentation site or official GitHub repository/discussion, checked directly on 2026-08-09. We did not use third-party blogs, aggregator listicles, or AI-generated summaries as evidence for any row. Where a framework's docs were silent, we searched its own GitHub issue tracker for the clearest available signal rather than reporting an absence as a guess.

Primary sources checked: [Google ADK A2A docs](https://adk.dev/a2a/); [Microsoft Agent Framework A2A integration](https://learn.microsoft.com/en-us/agent-framework/integrations/a2a); [Mastra A2A docs](https://mastra.ai/docs/agents/a2a); [Agno AgentOS A2A interface](https://docs.agno.com/agent-os/interfaces/a2a/introduction); [CrewAI A2A agent delegation](https://docs.crewai.com/en/learn/a2a-agent-delegation); [pydantic/fasta2a](https://github.com/pydantic/fasta2a); [LangChain LangSmith A2A server](https://docs.langchain.com/langsmith/server-a2a); [Microsoft AutoGen A2A discussion #6297](https://github.com/microsoft/autogen/discussions/6297); [OpenAI Agents SDK issue #1374](https://github.com/openai/openai-agents-python/issues/1374); [Letta documentation](https://docs.letta.com/); [Claude Agent SDK overview](https://code.claude.com/docs/en/agent-sdk/overview). This is a snapshot as of the checked date, not a standing guarantee; frameworks without support today may ship it tomorrow, and we'll re-check before restating any of these claims elsewhere on the site.

---

Tags: ai-agents, frameworks, a2a, mcp, integration, market-analysis, data-report