AI Agenter

    Connect an AI agent to Jira and Confluence in 30 minutes

    June 24, 2026·9 min read

    Practical mini-guide for product owners and scrum masters. GitHub Copilot CLI + Atlassian's MCP server (mcp-atlassian via uvx). Let a terminal-based AI agent read your Confluence docs, refine feature descriptions, and write directly back to your Jira tickets — gated by your "yes". Seven steps, ~30 minutes, Data Center and Cloud, PAT handling, READ_ONLY_MODE and project/space filters, troubleshooting and a hard security checklist at the end.

    > 📥 The full guide is available as a PDF (Danish, 10 pages). Download it at the top of the page — all 7 steps, ready-to-use commands, sample `mcp-config.json` for Data Center + Cloud, a troubleshooting table, and a security checklist.

    Your developers got faster with AI. Guess who's the bottleneck now. The PO or scrum master, sitting in Jira and Confluence trying to keep up with a team that suddenly ships twice as fast.

    This mini-guide gives you a terminal-based AI agent — GitHub Copilot CLI + the open-source mcp-atlassian server — that can read your Confluence, search your Jira issues, refine feature descriptions, and write back to tickets (after your "yes"). Total setup: about 30 minutes.


    The big picture

    You in the terminal → GitHub Copilot CLI (the agent you talk to) → MCP server (mcp-atlassian) (the bridge / translator) → Jira + Confluence (your data).

    • Copilot CLI is the agent.
    • MCP (Model Context Protocol) is an open standard that lets the agent use external tools.
    • mcp-atlassian is a ready-made MCP server that knows how to speak Jira and Confluence. You don't build the integration — you install and configure it.


    Setup in 7 steps (at a glance)

    1. Install Copilot CLI (WinGet or npm) → restart shell → `copilot --version`.

    2. First run + login. `copilot` from an empty workspace folder → trust → `/login`.

    3. Get your PATs. One in Jira, one in Confluence (Data Center). Cloud: API token at id.atlassian.com.

    4. Install uv / uvx. `winget install astral-sh.uv` → restart → `uvx --version`.

    5. Configure the MCP server in `~/.copilot/mcp-config.json` (Data Center: `JIRA_PERSONAL_TOKEN` + `CONFLUENCE_PERSONAL_TOKEN`; Cloud: `_USERNAME` + `_API_TOKEN`).

    6. Restart and verify. `/mcp` should show `mcp-atlassian` connected with tools like `jira_search` and `confluence_search`.

    7. Put it to work. Pull context from Confluence, refine features in a Jira project, and write back — the agent waits for your `yes` before changing anything.


    Three hygiene flags to set from day one

    • `READ_ONLY_MODE: "true"` on your first session. Confirm the agent can read, then flip to `"false"`.
    • `JIRA_PROJECTS_FILTER` and `CONFLUENCE_SPACES_FILTER` — scope the agent to specific projects/spaces.
    • Self-signed cert? Set `JIRA_SSL_VERIFY` / `CONFLUENCE_SSL_VERIFY` to `"false"` — but try without first.


    Security — do this once you're set up

    Tokens live in plaintext in `mcp-config.json`. Normal for MCP configs, but:

    1. Never commit `mcp-config.json` to Git, and never paste tokens in chats/screenshots.

    2. Set a sensible expiry on each token.

    3. Rotate immediately if you suspect exposure: create new, update config, revoke old.

    4. Use filters and read-only mode to constrain what the agent can touch.


    Bring an AI wingman

    Follow the steps in order — but you don't have to do it alone. Keep a strong model open beside you (e.g. Claude Opus 4.8 or Sonnet 4.6) as your wingman. When a command fails, paste the error and ask what went wrong. Have it translate a Windows step to macOS/Linux, explain a setting, or sanity-check your `mcp-config.json` before you save.


    Honest disclosure

    All commands tested on Windows 11 with GitHub Copilot CLI and the open-source mcp-atlassian server. The same steps work on macOS and Linux with small command tweaks. This is not an official Atlassian, GitHub, or Microsoft product.


    > ⚠️ Disclaimer. General practical guidance — not legal, security, or compliance advice. Giving an AI agent read/write access to Jira and Confluence touches your organization's data and credentials: clear it with your IT, security, and compliance teams, and follow your third-party tooling policy before enabling write mode in production. PATs act on your behalf — treat them as passwords. Tools, plans, pricing, and terms shift fast: verify against primary sources (GitHub Docs, github.com/sooperset/mcp-atlassian, your Atlassian admin). Snapshot as of June 2026.


    📬 Subscribe to the newsletter "AI, Built Human" on Substack — weekly insights on AI in practice.


    Stefano Vincenti · GenAI Strategist & Architect · External Lecturer, IT University of Copenhagen & DIS Copenhagen · Cofounder & CTO BotTellMe · Partner, TryZone · aitrainer.dk