Atlassian just opened what might be the most useful enterprise data graph in software to outside AI agents.
At Team '26 in Anaheim, the company announced that its Teamwork Graph — more than 150 billion objects and relationships spanning Jira, Confluence, Jira Service Management, and dozens of connected SaaS tools — is now available via MCP. Claude Code, IDE copilots, and any other MCP-compliant agent can now query the same context substrate that powers Atlassian's own Rovo AI platform.
"The context window is not stuffed anymore. You can actually use reasoning power where it belongs, not just to sit through a bunch of data."
— Jamil Valliani, Atlassian Head of Product for AI
What actually shipped
Teamwork Graph MCP servers (open beta) — exposes the graph to any MCP-compliant agent using Atlassian's internal query language, "Cipher," with multi-hop relationship traversal. Claude Code, Cursor, or whatever agent you're running can now query Jira history, Confluence decisions, and cross-tool relationships without custom glue code.
Teamwork Graph CLI (open beta) — a terminal interface to the graph for developers and admins. Free tier. CEO Mike Cannon-Brookes called this "probably going to be the number one thing received by customers."
Max — a new mode inside Rovo Chat, described as a "mini Claude Code" running in the cloud with Teamwork Graph context built in. Unlike Rovo's per-seat credit model, Max will bill on a variable consumption basis (rate cards coming "relatively soon").
teamworkgraph.com — a new public site that lets customers visualise their own graph for the first time.
Why graphs beat RAG here
This is the crux. The alternative to graph traversal is RAG — stuff a context window with retrieved documents, hope the model finds the right signal. Atlassian ran a benchmark comparing Claude Code with and without Teamwork Graph CLI access: 48% fewer tokens used, 44% more accurate results.
The mechanism is relational traversal rather than text retrieval. An agent reasoning over the graph can hop from a Jira ticket to the Confluence decision that motivated it, to the engineer who made the call, to the other tickets affected — without reconstructing that chain from raw text. That's genuinely different from dropping documents into a prompt.
The context layer is now the competition
Atlassian is explicit that this is a strategic bet. The argument: whichever platform provides the best context fabric for enterprise agents wins the agentic layer. And they're willing to open that graph even to agents that run on Anthropic's infrastructure rather than Atlassian's own Rovo.
They're not alone making this argument. Microsoft has Microsoft Graph + Copilot Studio. Salesforce has Data Cloud + Agentforce. ServiceNow announced competing tools at their own event the same week.
The differentiator Atlassian is betting on: years of structured relationship data across the tools teams actually use. Not a freshly built data layer — one that's been accumulating since the first Jira ticket.
What to do
- Using Claude Code or a similar coding agent? Try the Teamwork Graph CLI (open beta, free) and point it at your Atlassian workspace. The practical payoff is agents that understand which tickets and decisions are tied to the code you're editing.
- Evaluating MCP integrations? The Teamwork Graph MCP servers let you treat Atlassian as a context source without custom connectors.
- Building agentic tooling on Atlassian? Read the Cipher query docs — multi-hop traversal is now accessible to third-party agents, not just Rovo.
- Tracking the context-layer wars? Watch how Microsoft Graph, Salesforce Data Cloud, and Atlassian Teamwork Graph differentiate. This race determines where enterprise agents live.
Source: The New Stack — Why Atlassian is letting Claude Code into its own data graph
✏️ Drafted with KewBot (AI), edited and approved by Drew.
Top comments (0)