At Team '26 Europe in Amsterdam, Atlassian introduced AMP, the Agentic Multiplayer Protocol. The idea is simple and big: AI agents stop working one-on-one in hidden chats and start working with your team, in the open, inside Jira, Confluence and Loom.
From single-player AI to multiplayer teams
Most work with AI still happens in a private browser tab or a terminal session. The result is useful, but nobody else sees how it was made, and the context disappears when the tab closes. AMP sets the rules for agents to join the team instead.
Single-player AI
One person, one chat
- Work happens in private tabs and terminals
- Context is lost when the session ends
- Hard to see what an agent did, or who owns it
Multiplayer with AMP
Agents as teammates
- @mention agents in Confluence, Jira threads and Loom briefs
- Context comes from the Teamwork Graph
- Every agent has an owner, an identity and an audit trail
AMP in three parts
In the flow of work
Agents take part where work already happens: Confluence @mentions, Jira comment threads and Loom video briefs. The Atlassian MCP Server adds context from tools like Figma and your IDE.
Identity and presence
Each agent has a named owner and its own profile, and shows up in presence bars and cursors next to people. Local agent sessions on developer laptops appear on Jira boards.
Context and governance
Agents work from the Teamwork Graph within strict permissions. They act as the user or through a dedicated service account, and every action is logged.
“The best work has always been done in teams.” Mike Cannon-Brookes, CEO and Co-founder, Atlassian
The numbers behind the platform
How MCP and AMP fit together
Two protocols, two directions. MCP brings your Atlassian context into the AI tools you already use. AMP brings agents into your team's work, with identity, presence and governance.
Your AI clients
Claude, ChatGPT, Cursor, VS Code
Connected once, updated automatically
Atlassian MCP Server
215+ tools across 10+ Atlassian apps
Permissions, data policies, SSO
Teamwork Graph
Jira, Confluence, Loom, Bitbucket
Plus 80+ connectors and your code
Security teams keep control: admins choose what MCP can read, write and delete, allow or block AI access by site, space or classification, and run authentication through their own identity provider and SSO.
More from the Team '26 Europe keynote
- Agent Sessions brings cloud and local agent work into Jira and the Teamwork Graph, so context carries forward.
- Non-Human Identities gives every agent, Rovo or third-party, its own identity: what it can access and who owns it.
- Record for Agent turns a quick screen and voice recording into Jira work items, a prototype or a Confluence spec.
- Interactive PR Reviews has agents walk reviewers through their code changes in a Loom video, right in Bitbucket.
- Code Search, Atlassian Insights and Artifacts add source code, structured data and agent outputs to the Teamwork Graph.
How to get your team ready
- Tidy permissions and spaces. Agents respect the same access rules as people, so clean access makes safe agents.
- Decide who owns each agent. Choose between "run as user" and dedicated service accounts before agents go live.
- Connect your AI clients to the Atlassian MCP Server. Start with the tools your developers already use daily.
- Feed the Teamwork Graph. Add the connectors for the systems your teams rely on, so agents answer with full context.
- Start with one visible workflow. For example, an agent that drafts Jira updates in the open, reviewed by the team.
AMP in questions
What is Atlassian AMP?
AMP, the Agentic Multiplayer Protocol, is Atlassian's set of rules for AI agents working inside teams. It gives every agent an owner, an identity and a visible presence, grounds it in the Teamwork Graph, and lets people work with it in Jira, Confluence and Loom. Atlassian announced it at Team '26 Europe in October 2026.
How is AMP different from the Atlassian MCP Server?
They work in opposite directions. The Atlassian MCP Server brings Jira, Confluence, Loom and Bitbucket context into external AI tools such as Claude, ChatGPT, Cursor and VS Code. AMP brings AI agents into your team's own work, with identity, shared presence and governance.
How does AMP keep AI agents secure and governed?
Agents follow strict permissions and act either as the user or through a dedicated service account. Every action is recorded in an audit trail, and each agent has a named owner. For the MCP Server, admins control read, write and delete access, data policies by site or space, and sign-in through their own SSO.
Where can teams work with AI agents under AMP?
Where work already happens: @mentions in Confluence, Jira comment threads and Loom video briefs. Agents appear in presence bars and cursors next to people, and local agent sessions on developer laptops show up on Jira boards for the whole team.
How can my team prepare for AI agents in Atlassian?
Tidy permissions and spaces, decide who owns each agent, connect your AI clients to the Atlassian MCP Server, add connectors so the Teamwork Graph has full context, and start with one visible agent workflow that the team reviews together.
Bring agents into your team, the governed way
ONETEEM is an Atlassian partner building Rovo agents and Forge apps every day. We help you set up the Atlassian MCP Server, design agent ownership and permissions, and train your teams to work with agents in the open.
Sources: Atlassian press release, 7 October 2026 and Atlassian MCP.
