Jira’s depth, on demand: designing an MCP server around jobs, not endpoints

Atlassian’s Jira MCP server redesign moves beyond direct API calls to focus on user tasks such as closing issues or managing sprints, addressing real-world workflow complexities like validators, permissions, and post-functions that often cause API failures. By abstracting these details, the server handles coordination internally, allowing agents to prioritize outcomes over endpoint mechanics and reducing retries or errors tied to tenant-specific configurations.

The initial version mapped tools directly to REST APIs, requiring multiple calls for simple actions like creating an issue, which introduced inefficiencies and limited functionality. After internal debate, the team pivoted to a job-centric model, prioritizing common user tasks and consolidating related API interactions behind fewer, more purpose-built tools, resulting in 52 Jira tools that collectively call over 100 distinct APIs.

Security and governance remain central, with permissions enforced through OAuth 2.1 and additional controls available via Atlassian Guard’s Data Security Policies, allowing admins to restrict MCP server access to specific projects or sensitive content classifications. This ensures agents operate within the same boundaries as the signed-in user, preventing unauthorized actions even if the user has access in the browser.

The updated approach defines tools around concrete jobs like managing sprints or transitioning issues, where a single tool may coordinate multiple APIs behind the scenes. For example, a tool to close done sub-tasks and carry the rest into the next sprint handles issue transitions, sprint updates, and comment generation without requiring the agent to orchestrate each step individually.