Native Support for A2A Protocol

hello there!

what’s the plan from langchain to support a2a protocol natively? Agent2Agent (A2A) Protocol

langchain supports MCP adaptors with GitHub - langchain-ai/langchain-mcp-adapters: LangChain 🔌 MCP which translates MCP tool definitions to langchain @tool and also translates tool calls to MCP tool calls.

A2A is a also similar, but instead of tools, we now have agents. what’s langchain’s opinion on adding support for adding remote agents as sub-agent/remote-graph natively in langgraph.

I could see workarounds by creating Custom implementation and commands to intercept and execute remote agent calls.

looking forward to see a2a adopted natively here.

Hiya! I guess I have 2 responses for now.

We do have native A2A support in LangGraph via the deployments product: A2A endpoint in Agent Server - Docs by LangChain

As for the open source repo, we will scope out some more native ways to configure cards, skills, A2AUI components, etc. OSS doesn’t have a particular “server” implementation yet, and A2A as a protocol is still is over-engineered and under-specified, but we’ll work on making it easier to get started there.

hey @wfh , what do you mean by A2A as a protocol is over-engineered and under-specified?

could you elaborate?

in the current langgraph oss – there isn’t an easy way to delegate an execution from a node to a remote agent. what i’m interested to look on the A2A roadmap is

  • delegating to remote agents
  • human in the loop propagation from a remote agent to upstream caller
  • mechanism to send task-update-events from a langgraph node
  • a2ui is definitely a cleaner way to bring a2a and langgraph closer