Now GA: Connect any third-party AI agent to Rovo with A2A

[#1]

We're excited to announce Agent2Agent (A2A) connections are now generally available.

If you've already connected Claude, Cursor, or another AI tool using the Atlassian Rovo MCP server, you know how powerful it is to pull live Atlassian context into your workflow. A2A takes that further, instead of a single AI tool querying Atlassian, any compliant third-party AI agent can now connect directly to Rovo, reason over your real Atlassian data, and take action on your behalf.

What's now possible

Here's what this looks like in practice:

Your Gemini agent connects to Rovo via A2A and checks your Jira board to see which marketing announcements went out recently. It then pulls Google performance data to see how those posts are landing. From there, it drafts a Confluence page summarising the results, emails it to your team, and raises Jira tickets for the trends worth acting on — opportunities you haven't targeted yet but that are already gaining traction.

No tab-switching. No copy-paste. The agent closes the loop for you.

A2A is an open protocol, so this works with any A2A-compliant agent. To connect, point your agent at our agent card: https://a2a.atlassian.com/.well-known/agent.json.

Already using Gemini? Connect it to Rovo in one click via the Google Cloud Marketplace.

Get started

General documentation: Use Agent2Agent connections

Admin documentation: Manage A2A connections in Atlassian Administration

Join the webinar

Want to see it in action? Join us for "Open for Business: Connect any AI agent to Rovo with A2A." We'll walk through how it works and answer your questions live.

Register here: https://ace.atlassian.com/events/details/atlassian-atlassian-developers-presents-open-for-business-connect-any-ai-agent-to-rovo-with-a2a/

The A2A protocol is still new and we're building it in the open. Try it and let us know what you think in the comments.

[Matt Richards]

@Jemma Swaak - Can you provide a detailed comparison between the capabilities of your MCP and A2A connectors? 

[Jindong Li]

perhaps a little bit about mcp vs a2a protocol level difference...

[Jatin K Malik]

Hello @Matt Richards & @Jindong Li

Since you're both asking about MCP vs A2A, here's one answer covering capabilities and protocol.

---

Capabilities

• MCP is for AI clients like Claude, ChatGPT, or an IDE that want direct, granular access to Atlassian data. The client calls individual tools, like searchJiraIssuesUsingJql or createConfluencePage, and decides step by step what to call and when.
• A2A flips that around. An external agent hands Rovo a whole task in natural language, something like "summarize my open Jira bugs into a Confluence page", and Rovo works out the steps internally and hands back a result. Rovo currently advertises this as two "skills": Work in Jira and Work in Confluence, rather than a flat list of functions.

---

Protocol

• MCP runs on mcp.atlassian.com over Streamable HTTP (bidirectional), and supports OAuth 2.1 and API tokens. Discovery happens through OAuth Protected Resource Metadata plus a tools/list call.
• A2A runs on a2a.atlassian.com over JSON-RPC 2.0 via HTTP POST, with SSE for streaming methods like message/stream and tasks/resubscribe. Auth is OAuth 2.1 only, and discovery happens through a public Agent Card at /.well-known/agent.json instead of a tools list. Since A2A hands off whole tasks instead of individual calls, it also tracks task state and artifacts, with lifecycle methods like tasks/get and tasks/cancel.

---

Either way, both respect whatever Atlassian permissions the requesting user already has, so the security model doesn't change underneath you. The Agent Card itself is public at https://a2a.atlassian.com/.well-known/agent.json if you want to see what Rovo actually advertises.

Please let me know if you have any follow-up questions?