Atlassian Code Context: Give Your AI Agent Cross-Repo Vision

Atlassian's Code Context, now in open beta, indexes codebases into the Atlassian Teamwork Graph, giving AI coding agents cross-repo visibility via MCP-compatible clients. Internal testing showed agents with Teamwork Graph context produced 44% more accurate results while consuming 48% fewer tokens. The feature supports GitHub and Bitbucket repositories and works with Claude Code, Cursor, GitHub Copilot, Codex, Rovo Dev, and Rovo Chat.

Your AI coding agent knows your current repo. It has no idea about the three services it depends on, the Jira tickets explaining why the architecture looks the way it does, or the Confluence doc warning against the exact pattern it’s about to recommend. That is the multi-repo context wall — and it has been the quiet tax on AI-assisted development since agents became productive. Atlassian’s Code Context https://www.atlassian.com/blog/development/code-context , now in open beta, is a direct attack on it. The Problem Every Multi-Repo Team Hits AI coding agents in 2026 can see what’s in front of them. That’s it. Cross-repo dependencies get dropped entirely. Agents miss call chains between services because each repository gets indexed in isolation. Cursor’s local codebase index maxes out on 500K+ line projects and freezes. Even Claude Code, with its 200K token context window, cannot bridge multiple live codebases simultaneously. The community workaround — hand-maintained AGENTS.md context files — goes stale within days and costs developer time to keep current. This is the actual problem Code Context is solving. What Code Context Does Code Context is a new capability inside the Atlassian Teamwork Graph. It indexes your codebase — single repos or multi-repo setups — into the same graph that already tracks Jira issues, Confluence pages, pull requests, and team data. Once indexed, that codebase becomes queryable from everywhere Atlassian and its connected tools operate: your IDE, Jira, Rovo Chat, and any MCP-compatible AI client. The benchmark numbers from Atlassian’s internal testing are worth citing specifically: agents enriched with Teamwork Graph context produced 44% more accurate results while consuming 48% fewer tokens compared to agents without that context. The accuracy gain is expected. The token efficiency number is the one engineering leads should care about — at the scale of CI/CD pipelines running agent tasks daily, 48% fewer tokens is a material cost reduction, not a footnote. How It Works: The MCP Interface Code Context surfaces through two new Rovo MCP Server tools, which any MCP-compatible AI client https://support.atlassian.com/atlassian-rovo-mcp-server/docs/supported-tools/ can call: getTeamworkGraphContext — pass it a Jira issue, Confluence page, pull request, or any work item, and it returns everything connected to that object in the Teamwork Graph: linked issues, related documentation, PRs, deployments, collaborators, and now source code across all indexed repos. getTeamworkGraphObject — takes the connections returned above and fetches their full content. These two tools work as a discover-then-fetch pair. The agent maps what’s connected first, then reads what it needs. This is more token-efficient than stuffing everything into a prompt upfront — which is where the 48% savings largely comes from. Permission scoping is built in: agents only retrieve code and context that the requesting user already has access to through existing SCM permissions. No broad access grants required. Who Can Use It Now Code Context is in open beta. It supports GitHub and Bitbucket repositories at launch. Enabling it requires an org admin to opt in through Atlassian Administration under Rovo settings — once enabled, code indexing runs in the background and becomes queryable immediately. On the client side, compatibility is broad: Claude Code, Cursor, GitHub Copilot in VS Code, Codex, Rovo Dev, and Rovo Chat all work via MCP. You are not required to use Atlassian’s own tooling to benefit from this — any agent that speaks MCP can now query your organization’s codebase through the Teamwork Graph. The Rovo Dev extension for VS Code https://www.atlassian.com/software/rovo-dev makes the connection straightforward for most teams. The Strategic Move Underneath the Feature Atlassian’s competitive position rests on data connectivity. It has years of Jira history, Confluence documentation, and organizational work data that no standalone AI coding tool can replicate. Cursor does not know your backlog. GitHub Copilot does not understand why you made an architectural decision three sprints ago. By making Code Context available over MCP — so that Claude Code and Cursor can query it — Atlassian is not just adding a feature. It is positioning the Teamwork Graph as the authoritative context provider for enterprise development, regardless of which AI coding tool a team uses. Teams stay in their preferred IDE. The Teamwork Graph provides the organizational memory. That is a sensible moat to build in 2026, and Code Context is the concrete implementation of it. The open beta is accessible now through Atlassian Administration. For teams running multi-repo codebases with Jira and Bitbucket or GitHub, this is worth enabling this week. For more on the multi-repo agent memory problem Code Context addresses, this breakdown of the bottleneck https://supermemory.ai/blog/memory-bottleneck-large-repo-coding-agents/ is useful background.