Atlassian’s Teamwork Graph update is best understood as context infrastructure for AI agents. At Team ’26, Atlassian said it is opening the graph so Rovo and agents from the broader ecosystem can search, reason, and act across tools and teams ^[9]. The two new access paths are Teamwork Graph tools exposed through Rovo’s Model Context Protocol server and a Teamwork Graph CLI, both described as beta or open-beta interfaces in reporting and Atlassian materials ^[1][3][7].

The promise is simple: instead of forcing an AI model to process broad piles of enterprise content, agents can retrieve more specific work context from a graph that already maps teams, projects, documents, decisions, and related work. Atlassian says its benchmarks found responses grounded in Teamwork Graph data were 44% more accurate while using 48% fewer tokens ^[7].

The two new access paths

Teamwork Graph tools through Rovo’s MCP server. Atlassian is delivering Teamwork Graph access through Rovo’s Model Context Protocol server, giving Rovo and third-party AI agents more fine-grained access to graph context ^[1][3]. Atlassian says agents can pull connected context from Jira, Confluence, Jira Service Management, Loom, and integrated third-party tools ^[6].

Teamwork Graph CLI. The command-line interface gives developers and coding agents a terminal-friendly way to explore and query Teamwork Graph context ^[1][7]. SiliconANGLE reported that the CLI includes more than 300 commands and can let coding agents such as Claude Code and Cursor query work and relationships in the graph ^[3].

Together, the MCP path and CLI move Teamwork Graph beyond Atlassian’s own AI surfaces. The MCP interface is the agent-facing route; the CLI is the developer and workflow route.

What the Teamwork Graph adds

Atlassian describes Teamwork Graph as a shared context layer for how work happens across an organization. Coverage of the launch says the graph connects people, projects, documents, decisions, and work across Atlassian and third-party tools, and now contains more than 150 billion connections ^[3]. Atlassian’s own product page frames it as a way to unify work data for contextualized AI experiences and cross-app visibility ^[6].

That matters because enterprise AI agents often fail when they lack organizational context. A model may understand language, but it still needs to know which project, goal, owner, document, dependency, or decision is relevant. Teamwork Graph is Atlassian’s attempt to provide that context as a structured layer rather than as a giant prompt.

How the token savings are supposed to work

The token-cost argument comes down to retrieval precision. TechTarget reported that Atlassian’s new MCP and CLI tools give agents more fine-grained access to Teamwork Graph context and reduce noisy data exchange among agents ^[1]. If an agent can fetch only the relevant linked work, decision, document, or owner, it does not need to send as much irrelevant text into the model.

Atlassian’s benchmark claim is that grounding responses in Teamwork Graph data delivered 44% more accurate results while using 48% fewer tokens ^[7]. TechTarget also characterized the update as targeting token costs, reporting potential reductions of up to 48% ^[1]. The figure should be read as a vendor benchmark, not a guaranteed discount for every deployment.

In practice, savings depend on how teams wire the tools into agents, which models they use, how much context agents retrieve, and whether their connected work data is complete enough. The architectural shift is still important: Atlassian is trying to replace broad context stuffing with narrower, relationship-aware context retrieval.

Why enterprises should care

For enterprises, the update is not just about giving agents more data. It is about giving them more selective data. Atlassian says the opened Teamwork Graph is intended to let Rovo and ecosystem agents search, reason, and act securely across tools and teams ^[9]. The product page also emphasizes connecting agents to Atlassian context across browser, desktop, and terminal workflows ^[6].

That could make external agents more useful in software development, service management, knowledge work, and cross-functional planning, especially where the answer depends on relationships between Jira issues, Confluence pages, ownership, goals, and decisions. The CLI is particularly relevant for developer workflows, while the MCP server gives agent platforms a standardized way to request context ^[3][7].

What to watch in beta

Because the MCP tools and CLI are still beta or open beta, enterprises should validate the claims in their own environment before assuming large cost savings ^[1][3][7]. The key questions are whether the graph has enough coverage, whether permissions and security controls match enterprise requirements, and whether token telemetry shows prompts getting smaller without making answers worse.

The bottom line: Atlassian launched Teamwork Graph access through Rovo’s MCP server and a new Teamwork Graph CLI. The economic bet is that agents grounded in a structured work graph can retrieve better context with fewer tokens than agents relying on broad prompt stuffing.