Jira is a widely used project management platform developed by Atlassian. It allows teams to plan, track, and manage work using issues — tasks, bugs, stories, and epics — organised into boards and workflows. At bacher-ai, Jira is used to track all platform development work. And now, Claude can manage it directly through conversation.
Project management is one of those tasks that always competes for attention. Logging a ticket, updating its status, adding a comment — each step is a small friction point that interrupts focused work. The new Jira MCP server eliminates that friction entirely, letting Claude manage issues through natural language.
From Conversation to Ticket
The idea is straightforward. Instead of switching to a browser, navigating to Jira, and filling in forms, you simply tell Claude what you need. “Create a task for the logo redesign, due Sunday.” “Mark BAI-26 as done and add a comment about what was implemented.” “What’s overdue on my board?” Claude handles the rest — selecting the right tool, constructing the correct API call, and confirming the result.
This is made possible through the Model Context Protocol (MCP), which allows AI assistants like Claude to call external tools as part of a conversation. The Jira MCP server on bacher-ai is built in n8n and exposes eight tools covering the full lifecycle of an issue: listing, searching, reading, creating, updating, transitioning, commenting, and deleting.
Built on the Jira REST API
The first version of the Jira MCP server used n8n’s native Jira nodes. It worked — but it had bugs. Labels were passed as strings instead of arrays, status transitions hit the wrong endpoint, and optional fields were incorrectly marked as required. These weren’t minor annoyances; they made the tool unreliable in practice.
The fix was to bypass the native nodes entirely and call the Jira REST API v3 directly using n8n’s HTTP Request tool nodes. This gives full control over the JSON payload. Labels are sent as proper arrays. Transitions use the dedicated /transitions endpoint. Optional fields are omitted cleanly using an arrow function pattern that drops undefined values before the request is serialised. The result is a tool that behaves exactly as intended every time.
Skill-Guided Tool Use
Having the MCP tools available is only half the story. The other half is making sure Claude uses them correctly and consistently. For that, bacher-ai uses a skill file — a structured markdown document that Claude reads before handling any Jira-related request.
The skill defines which tool to use in which situation, what the default values are (project, assignee, priority), how the issue type hierarchy works, and a set of workflow rules: check for duplicates before creating, transition to In Progress when starting work, add a closing comment when marking something Done. These rules encode the project’s conventions directly into Claude’s behaviour, so the same good habits are applied every session without needing to be re-explained.
Tested Thoroughly
All eight tools were validated across multiple sessions covering the full issue lifecycle: tasks created, read back, updated with new fields, commented on, and transitioned through all three statuses. Subtask creation with parent linking was verified against the Jira web interface. Each tool was exercised repeatedly in realistic scenarios — not just isolated tests, but actual project management work: closing completed issues with implementation notes drawn from Nextcloud, reviewing the board for upcoming deadlines, and tracking work in progress.
The Jira MCP server is one piece of a broader vision at bacher-ai: an AI assistant that can help manage real work, not just answer questions about it.
Key Takeaways
- The bacher-ai Jira MCP server exposes 8 tools covering the full issue lifecycle, callable by Claude through natural conversation.
- Rebuilding on raw HTTP Request nodes instead of n8n’s native Jira nodes fixed four bugs and gave full control over the API payload.
- A skill file encodes project conventions into Claude’s behaviour — default values, tool selection rules, and workflow habits applied consistently every session.
- The tool was validated thoroughly across multiple sessions, including subtask parent linking and real project management work.
Photo: cottonbro studio via Pexels