Atlassian.Mcp.Server 0.1.6-prerelease.gd2996b2

Atlassian MCP Server

A Model Context Protocol (MCP) server that lets AI assistants such as Claude Code and GitHub Copilot work with Jira Cloud and Confluence Cloud. It runs on your machine, talks to your site with your own API token, and needs no Rovo credits.

This is an unofficial, community project. It is not affiliated with or endorsed by Atlassian.

Features

Jira

• Issues: get, create, update, delete, assign, transition, labels, watchers, votes, and change history.
• Custom fields: set any field, including custom fields, and discover a project's required fields and allowed values before creating an issue.
• Markdown everywhere: descriptions, comments, worklog comments, and multi-line custom fields are written in Markdown and converted to Atlassian Document Format; rich text is read back as Markdown.
• Search: JQL search with paging, approximate totals, JQL validation, and value suggestions.
• Comments, links, and attachments: read and write comments, link issues by link type name or phrase (such as "is blocked by"), and upload, download, or delete attachments.
• Projects and people: projects, components, versions, users, groups, and permissions.
• Time tracking: worklogs and estimates.
• Filters: saved filters and favorites.
• Agile: boards, backlogs, sprints, epics, ranking, and sprint report, burndown, and velocity data.

Confluence

• Private drafts: new pages and changes to existing pages are saved as drafts by default, so you can preview them in Confluence before anything is published.
• Safe editing: edit one section of a page and keep everything else, including macros and smart links; edits that would lose content Markdown cannot represent are refused.
• Pages: read pages as Markdown or ADF, create, update, publish, move, and delete them, and browse children and version history.
• Search: CQL search across spaces.
• Spaces, comments, attachments, and labels: upload, list, and delete attachments (deleting moves an attachment to the space's trash, and needs the opt-in confluence-admin toolset).

Tools are grouped into toolsets so you can enable only what you need. See the full list of tools.

System requirements

• Windows, macOS, or Linux.
• The .NET 10 SDK or later, which provides the dnx and dotnet tool commands used to run the server.
• An MCP client, such as Claude Code, GitHub Copilot in Visual Studio Code or Visual Studio, or the GitHub Copilot CLI.

Atlassian requirements

• A Jira Cloud site, a Confluence Cloud site, or both, such as https://example.atlassian.net.
• An Atlassian account on that site.
• An API token for that account. See Manage API tokens for your Atlassian account.

The server acts as you: it can see and change only what your account can, and every change appears in Jira and Confluence history under your name.

Getting started

1. Create an API token at https://id.atlassian.com/manage-profile/security/api-tokens.
2. Add the server to your AI assistant. Every client runs the same command, dnx Atlassian.Mcp.Server --yes, which downloads the server from NuGet and runs the latest version each time it starts. Follow the guide for your assistant:
   • Claude Code
   • GitHub Copilot in Visual Studio Code
   • GitHub Copilot in Visual Studio
   • GitHub Copilot CLI If your workspace has a global.json that pins an older .NET SDK, dnx cannot run there, because clients start the server in the workspace folder. Install the server as a global tool instead, from a folder without that global.json: dotnet tool install --global Atlassian.Mcp.Server. Then use atlassian-mcp-server as the command, and update it with dotnet tool update --global Atlassian.Mcp.Server.
3. Try it. Ask your assistant, for example:
   • "Show me my open Jira issues in project PROJ."
   • "Create a bug in PROJ for the login timeout, with steps to reproduce."
   • "Draft a Confluence page in my personal space summarizing PROJ-123." Then open the link it returns to preview the draft.

Configuration

The server is configured with environment variables, which each guide shows how to set.

Uploading files by path

The attachment tools accept a file as base64 or as the path of a local file. Use a path for anything already on disk, such as a screenshot or a log: base64 makes the model copy hundreds of kilobytes of text character for character, and one wrong character uploads a corrupted file.

A path lets the model send a file from your computer to your site, so the server decides which files it may read, and asks you rather than the model:

• A file in an allowed folder is uploaded without asking. The allowed folders are those in ATLASSIAN_ATTACHMENT_FOLDERS, plus those you chose to always allow, which the server saves in %APPDATA%\Atlassian.Mcp.Server\attachment-folders.json (~/.config/Atlassian.Mcp.Server/attachment-folders.json elsewhere). Remove a folder by editing that file; the change applies immediately, in every client and session.
• For any other file, the server asks you, through your MCP client, whether to upload it, showing the full path, its size, and where it goes: Allow once, Always allow its folder, or Deny. This uses MCP elicitation, which Claude Code and GitHub Copilot in VS Code support. A question not answered within two minutes counts as Deny.
• In a client that cannot ask, such as Claude Desktop, a file outside the allowed folders is refused.
• Relative paths, files over 100 MB, and files reached through a symbolic link or junction are always refused. Folders are compared as written: a Windows 8.3 short path such as C:\Users\FIRSTN~1\… does not match the same folder written in full.

List or always allow only folders whose files you are willing to share, such as a screenshots folder. atlassian_jira_get_configuration shows both lists.

Some clients limit how many tools they send to the model; GitHub Copilot allows about 128. The guides show a toolset selection that stays under that limit.

More information

• Tools
• Contributing
• Changes

License

MIT