Model or dataset
sooperset/mcp-atlassian avatar
sooperset/mcp-atlassian

mcp-atlassian puts 98 Jira and Confluence tools behind one uvx command

MCP server for Atlassian tools (Confluence, Jira)

6,122 stars1,459 forksPythonMIT

At a glance

What is it?
mcp-atlassian is a Python MCP server that gives an assistant 98 tools for Jira and Confluence, launched through uvx. It covers Atlassian Cloud and Server or Data Center installs, and the two deployment types want different credentials.
Who is it for?
Choose it if you want an assistant to search and edit Jira and Confluence over the Model Context Protocol, and you have either an Atlassian Cloud API token or a Server or Data Center personal access token in hand. Look elsewhere if your instance sits below Jira 8.14 or Confluence 6.0, and go to the HTTP Transport documentation if you want a shared or hosted setup, because the quick start in the repository configures a local process only.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository received new commits within the last day.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • sooperset/mcp-atlassian6,122 ★ · +233
Star history
Repository2026-09-112026-11-15
sooperset/mcp-atlassian5.9k6.1k

233 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

One mcpServers block carries both products, and Cloud tokens differ from Server tokens

The quick start asks for one set of values per product and puts all six into a single mcpServers entry. The command is uvx with the argument mcp-atlassian, so the IDE starts the server on demand instead of installing anything into a Python environment first. The env block carries JIRA_URL and CONFLUENCE_URL plus a username and an API token for each product, which is the Cloud shape. Tokens are created on your Atlassian account page for API tokens. The same block changes shape on Server and Data Center: JIRA_PERSONAL_TOKEN takes the place of the username and API token pair, so one variable does the work of two.

json
{
  "mcpServers": {
    "mcp-atlassian": {
      "command": "uvx",
      "args": ["mcp-atlassian"],
      "env": {
        "JIRA_URL": "https://your-company.atlassian.net",
        "JIRA_USERNAME": "[email protected]",
        "JIRA_API_TOKEN": "your_api_token",
        "CONFLUENCE_URL": "https://your-company.atlassian.net/wiki",
        "CONFLUENCE_USERNAME": "[email protected]",
        "CONFLUENCE_API_TOKEN": "your_api_token"
      }
    }
  }
}

The .env file resolves credentials by walking five methods in a fixed order

The .env.example file spells out how the server picks credentials, and it does it as a five step precedence list. First OAuth Standard, taken when a client ID and secret are set, with Cloud also needing a cloud_id and Data Center not needing one. Second OAuth BYOT, taken when an access token is set, again with cloud_id for Cloud. Third username plus API token over basic auth, which the file calls the recommended method for Cloud. Fourth a personal access token, and this step is conditional: it is reached only when the URL looks like Server or Data Center and the PAT is set. Fifth ATLASSIAN_OAUTH_ENABLE=true, a minimal BYOT mode where tokens arrive per request. That fourth condition is the trap. JIRA_URL decides whether a personal access token is consulted at all, so a mistyped instance URL quietly changes which credentials go out. The same file also asks you not to wrap values in double quotes unless the value itself contains spaces.

uvx mcp-atlassian is the only command the quick start actually gives you

Two commands appear in the quick start, and neither one is pip. The first is the uvx entry inside the IDE configuration, which fetches and runs the published package on each launch. The second is the Autohand Code path, a single command that registers the same uvx server together with your Atlassian credentials, and it takes --scope project after add when you want the registration kept inside the current project instead of somewhere wider. The documentation index lists four installation routes under Installation: uvx, Docker, pip, and from source, but only uvx and the Autohand command are written out here, so pip and Docker readers get their directions from the site. The manifest does make a pip install useful on its own: [project.scripts] maps mcp-atlassian to mcp_atlassian:main, which gives you a console entry point carrying the same name as the uvx package. Python 3.10 or newer is required.

bash
autohand mcp add mcp-atlassian env \
  JIRA_URL=https://your-company.atlassian.net \
  [email protected] \
  JIRA_API_TOKEN=your_api_token \
  CONFLUENCE_URL=https://your-company.atlassian.net/wiki \
  [email protected] \
  CONFLUENCE_API_TOKEN=your_api_token \
  uvx mcp-atlassian

JQL on the Jira side, CQL on the Confluence side, 98 tools in total

The tool surface is split by product and by query language. On the Jira side jira_search runs JQL, jira_get_issue returns issue detail, jira_create_issue and jira_update_issue write, and jira_transition_issue moves an issue through its workflow. On the Confluence side confluence_search runs CQL rather than JQL, confluence_get_page and confluence_update_page handle page content, confluence_create_page makes pages, and confluence_add_comment appends comments. Those ten named tools are a sample out of 98 in total, so the quick tour is not a catalogue and the full set sits behind the tools reference page. The start using step shows the shape of a request: find issues assigned to me in PROJ project, search Confluence for onboarding docs, create a bug ticket for the login issue, update the status of PROJ-123 to Done. Each of those reads or writes a live instance with the credentials from the step before.

Server and Data Center support stops at Confluence 6.0 and Jira 8.14

Deployment type is the fork in the road, and it brings a version floor with it. Confluence Cloud and Jira Cloud are marked fully supported. Confluence Server and Data Center are marked supported from version 6.0, and Jira Server and Data Center from version 8.14. Anything below those numbers sits outside what the project claims, which is worth settling before you point a server at an instance that has gone years without an upgrade. Two further boundaries stand next to that table. The project is MIT licensed and says plainly that it is not an official Atlassian product, so there is no support agreement from the vendor and no compatibility promise beyond the table itself. And the two deployment types expect different credentials by design: Cloud pairs a username with an API token, Server and Data Center expects a personal access token instead.

The container is alpine, runs as a non-root user, and always pulls the wpad extra

The Dockerfile is a two stage build that spends most of its lines on reproducibility and on not running as root. The first stage starts from the Astral uv image on python3.13-alpine, sets UV_COMPILE_BYTECODE so bytecode is written at install time, and sets UV_LINK_MODE to copy because the uv cache is a mounted volume. A build argument named VERSION exists so the reported version can be patched when the image is built without a .git directory: the Dockerfile rewrites the fallback-version entry in pyproject.toml when that argument looks like a number. Dependencies come from a generated lockfile, first uv lock and then uv sync frozen with the wpad extra, no dev group and no editable install, and the project itself is synced in a second pass after the source is copied. Cached bytecode is then deleted out of the virtual environment. The final stage is plain python:3.13-alpine, adds a user named app, switches to it, and copies only the virtual environment across with ownership set to that user. Notice that the sync asks for the wpad extra unconditionally, so pypac lands in the container whether you use it or not, and that the image runs Python 3.13 while the package floor is 3.10.

uvicorn and starlette in the dependency list explain the HTTP transport page

Two dependencies point at an option the quick start never shows. uvicorn and starlette sit in the main dependency list rather than behind an extra, and the documentation index carries an HTTP Transport page covering SSE, streamable-http, and multi-user. That pairing is what a networked server needs. The README's quick start registers a local process that the IDE starts and talks to, and no port, host, or user separation appears in its configuration example. So the HTTP transport and the multi-user arrangement are documented on the site, while the quick start shows only the single user local case, with helm and smithery.yaml sitting at the top level as further deployment files.

Editorial conclusion

Choose it if you want an assistant to search and edit Jira and Confluence over the Model Context Protocol, and you have either an Atlassian Cloud API token or a Server or Data Center personal access token in hand. Look elsewhere if your instance sits below Jira 8.14 or Confluence 6.0, and go to the HTTP Transport documentation if you want a shared or hosted setup, because the quick start in the repository configures a local process only. Before you install, check which tools you actually need, since the README names ten of the 98 and points the rest at the tools reference page, and keep an eye on the dependency ranges, which cap both mcp and fastmcp below their next major versions.

Frequently asked questions

What does Atlassian MCP do?

It bridges Atlassian products, Jira and Confluence, with AI language models over Anthropic's MCP specification. It exposes 98 tools, including jira_search with JQL and confluence_search with CQL, so an assistant can find issues, read and write pages, and move an issue through its workflow.

how to install mcp-atlassian

Add it to your Claude Desktop or Cursor MCP configuration with the command uvx and the argument mcp-atlassian, then supply JIRA_URL, JIRA_USERNAME, JIRA_API_TOKEN, CONFLUENCE_URL, CONFLUENCE_USERNAME, and CONFLUENCE_API_TOKEN in the env block. Server and Data Center users swap the username and token pair for JIRA_PERSONAL_TOKEN. The documentation site also covers Docker, pip, and from source.

what is mcp atlassian

It is an open source MCP server written in Python for Confluence and Jira, supporting both Cloud and Server or Data Center deployments. It is MIT licensed and is not an official Atlassian product. Server and Data Center support starts at Confluence 6.0 and Jira 8.14.

mcp atlassian vscode

The quick start shows configuration for Claude Desktop and Cursor, not for VS Code. The documentation index has a Configuration page covering IDE setup and environment variables, and an HTTP Transport page covering SSE, streamable-http, and multi-user, which is where a shared or hosted setup is described.

How to use Atlassian MCP?

Ask the assistant in plain language: find issues assigned to me in PROJ project, search Confluence for onboarding docs, create a bug ticket for the login issue, or update the status of PROJ-123 to Done. Each request runs against the instance whose URL and credentials you put in the configuration.

Official sources

  1. License: MIT
  2. Project website
  3. README
  4. Releases
  5. sooperset/mcp-atlassian on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/sooperset-mcp-atlassian.svg)](https://hysenlabs.com/projects/sooperset-mcp-atlassian)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/sooperset-mcp-atlassian.svg)](https://hysenlabs.com/projects/sooperset-mcp-atlassian)

Was this useful?