A2A vs MCP
MCP connects an AI agent to tools and data. A2A connects one agent to other agents. They are built to work together.
MCP and A2A solve different problems, and the A2A docs say they are not competitors. MCP is how an agent uses a tool, such as a database or an API. A2A is how one agent hands work to another agent, even one built by another team or company.
Our advice: if you only want Claude, ChatGPT or Cursor to reach your apps, you need MCP, not A2A. If you are building several agents that must cooperate, you will likely use A2A between them and MCP inside each one.
Good for
- Developers building agents that must work with other teams' agents
- People who use Claude or ChatGPT and keep hearing about both protocols
- Architects choosing standards for an agent platform
Side by side
Our advice: Use MCP to give one agent its tools. Use A2A when separate agents must discover each other and share a task.
| Option | MCP | A2A |
|---|---|---|
| Connects | Agent to tool or data | Agent to agent |
| Docs call it | Vertical | Horizontal |
| Other side is | Tool, API or database | Opaque remote agent |
| Typical call | Run one tool | Hand off a task |
| Discovery | Server lists its tools | Agent Card (JSON) |
| Long-running work | Tasks extension (poll for status) | Tasks, streaming, webhooks |
| Wire format | JSON-RPC 2.0 | JSON-RPC, gRPC or HTTP+JSON/REST |
| Runs over | stdio or HTTP | HTTP |
| Seen in | Claude, ChatGPT, Cursor | LangGraph, CrewAI |
| Governance | LF Projects, LLC series | Linux Foundation |
Which to pick
Our advice: Use MCP to connect an agent to its tools and data. Add A2A only when separate agents, possibly from different teams or vendors, must hand tasks to each other. The A2A docs recommend using both: A2A between agents, MCP inside each.
The A2A docs say it is not a replacement for MCP, and MCP is not an agent-to-agent protocol.
Best for
- MCP: connecting Claude, ChatGPT or Cursor to Notion, GitHub or a database
- MCP: giving one agent a typed list of tools it can call
- A2A: agents built on different frameworks that must delegate tasks to each other
- A2A: long-running tasks that stream progress or notify a client when done
Skip it if
- MCP: you expect it to coordinate with another team's agent
- A2A: you only need one assistant to reach your apps (our advice)
Related plugs
6
Notion MCPOfficialHostedOAuthNotion's hosted remote MCP server at mcp.notion.com. Sign in with OAuth and an AI client can search, read, create and update the pages and databases you can access. No local install.
GitHub MCP ServerOfficialHosted or LocalOAuthGitHub's official MCP server. Lets an AI assistant read code, manage issues and pull requests, check Actions runs and security alerts. Hosted at api.githubcopilot.com/mcp or run locally with Docker or a Go binary.
34.1K GitHub stars
LinearOfficialHostedOAuthFind and manage issues, projects, and documents in your Linear workspace.
SupabaseOfficialHosted or LocalOAuthMCP server for interacting with the Supabase platform
3K GitHub stars
Playwright MCPOfficialLocalPlaywright MCP server
38.9K GitHub stars
Context7OfficialHosted or LocalAPI keyUp-to-date docs for AI agents
64K GitHub stars
Questions
Is A2A a replacement for MCP?
No. The A2A docs say it is not a replacement for MCP. MCP standardizes how an agent connects to tools. A2A standardizes how agents talk to each other.
What is the difference between A2A and MCP?
MCP connects an agent to tools, APIs and data. A2A connects an agent to other agents. The A2A docs describe MCP as vertical, deepening one agent, and A2A as horizontal, reaching across teams or companies.
Can I use A2A and MCP together?
Yes, and the A2A docs recommend it. In their example, an auto-repair shop agent talks to a customer and a parts-supplier agent over A2A, while its mechanic agent uses MCP to reach diagnostic tools and a repair manual.
Do I need A2A to connect Claude or ChatGPT to my apps?
No. Connecting an assistant to apps like Notion or GitHub is what MCP does. A2A matters when you build or run agents that must delegate work to other agents. See what MCP is.
Who created A2A and MCP?
A2A was originally developed by Google and donated to the Linux Foundation, under the Apache 2.0 license. Its steering committee includes AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP and ServiceNow. MCP is documented as an open-source standard at modelcontextprotocol.io.
How does an agent find another agent in A2A?
Through an Agent Card, a JSON file that describes the agent's identity, skills, endpoint and sign-in requirements. A client reads the card, then sends a task.
How do MCP and A2A send messages?
MCP uses JSON-RPC 2.0 over stdio for local servers or streamable HTTP for remote ones. A2A runs over HTTP with JSON-RPC, gRPC and HTTP+JSON/REST bindings, plus streaming with server-sent events and push notifications for long tasks.
Can an A2A agent look like an MCP tool?
Sometimes. The A2A docs say a remote agent can expose well-defined, stateless skills as MCP-compatible resources. They add that A2A's strength is stateful, collaborative work that goes beyond a tool call.
Is A2A about an agent's own sub-agents?
No. The A2A docs say it does not specify how an agent talks to its own sub-agents or invokes tools. Use your framework's own features, or MCP, for those.
Sources
Docs, issue trackers and discussions this page is built from.
Show all 7Hide
- A2A Protocol · a2a-protocol.org
- A2A and MCP · a2a-protocol.org
- What is the Model Context Protocol (MCP)? · modelcontextprotocol.io
- MCP architecture overview · modelcontextprotocol.io
- Core Concepts · a2a-protocol.org
- A2A specification · a2a-protocol.org
- Governance and Stewardship · modelcontextprotocol.io