Atlassian launches AMP: the Agentic Multiplayer Protocol that gives AI agents an identity on your team
Atlassian unveiled AMP — the Agentic Multiplayer Protocol — at Team '26 Europe in Amsterdam on October 7: a framework that gives AI agents an assigned identity, scoped authority and shared Teamwork Graph context, plus a rebuilt Atlassian MCP server with 220+ tools, now generally available.
Atlassian used its Team ’26 Europe keynote in Amsterdam on October 7 to unveil AMP — the Agentic Multiplayer Protocol — a framework designed to turn AI agents from invisible helpers into visible, accountable collaborators inside Jira, Confluence and the rest of its system of work. Alongside AMP, the company announced a rebuilt Atlassian MCP server, enhancements to the Teamwork Graph, and new capabilities for customers to collaborate with agents in richer, more visible ways.
Agents operating under AMP get four things today’s enterprise agents mostly lack, per the company’s October 7 announcement: an assigned identity, scoped authority, shared context drawn from the Teamwork Graph, and reviewable output. The point is legibility: anyone on a team should be able to see what an agent did, under whose authority, and with what context — and check the work afterward.
Agents as accountable teammates
The mechanics build on what Atlassian has been rolling out all year: teams can assign work items to agents, @mention them in comments, and embed them directly into Jira workflows and automations, turning agents from scattered experiments into accountable teammates. Jira agent sessions tie local and cloud agent runs back to work items, so the record of who — or what — did what stays in the system of record rather than scattered across chat logs.
The scale is already large. According to coverage of the announcement, humans and agents now collaborate more than 10 million times a month across Atlassian’s products — which is exactly why the company wants a protocol for it rather than ad-hoc tooling.
The rebuilt MCP server
The other headline is the Atlassian MCP server, now generally available after a full rebuild: from dozens of tools to 220+ tools, secured with OAuth 2.1 + PKCE by default and tiered read/write/destructive scopes. Atlassian says the server already handles 15M+ calls a day, and that the rebuild cuts token use by up to 25% on the same Jira/Confluence work in internal Claude benchmarks. EU-local AI inference has been added, and a non-human identity inventory — a catalog of every agent operating in your environment — is coming soon.
A bigger Teamwork Graph
Underpinning both AMP and the MCP server is the Teamwork Graph, Atlassian’s shared context layer, which the company says now holds 250B+ connections and — new with this release — indexes Bitbucket and GitHub source code down to functions, symbols and classes through Rovo Code Search, no clone required. Atlassian’s internal benchmarks claim agents grounded in the graph are 44% more accurate while using 48% fewer tokens.
Why identity is the real product
The enterprise AI bottleneck has shifted: model smarts are cheap, coordination and accountability are not. AMP reads as Atlassian’s bid to make Jira the system of record not just for human work but for agent work too — with identity, scoping and reviewability as the governance layer. It lands the same week as the Personal Agent Protocol from Sierra, Meta, Walmart, Stripe and Shopify: protocols, suddenly, are the enterprise AI battleground.
Context matters here. Just a day earlier, on October 6, Atlassian expanded its partnership with OpenAI to bring frontier GPT-6 models into its platform and the Rovo assistant, per analyst coverage. AMP is the control plane being built on top of a much bigger model push — and the release roundup notes the pieces are designed to fit together: agents with identities, tools via MCP, and the graph as shared memory.
What to watch
The announcement leaves open questions worth tracking: whether AMP will be an open specification other vendors can implement or an Atlassian-platform framework, how it will be priced and licensed, and when the non-human identity inventory ships — that’s the piece auditors and security teams will ask about first. If Atlassian can make agent activity as legible as a Jira ticket, it owns the enterprise’s agentic audit trail. If it can’t, AMP is just a brand name for features competitors already ship.