🔗 Atlassian MCP Server

An MCP (Model Context Protocol) server that gives AI assistants full access to Jira, Confluence, and Bitbucket Cloud APIs -- 102 tools covering deep CRUD across all three products.

What It Does

This server acts as a bridge between any MCP-compatible AI client (Claude Desktop, etc.) and your Atlassian Cloud instance. Instead of switching between browser tabs, you ask your AI assistant to search Confluence, create Jira tickets, review pull requests, and more -- all through natural language.

Key Features

• Confluence (22 tools) -- Search (text or raw CQL), create, read, update, and delete pages. Manage spaces, child pages, comments (full CRUD, replies, and text-anchored inline comments), attachments, labels, and version history. Input accepts Markdown (auto-converted to ADF); output is native ADF JSON.

• Jira (50 tools) -- Full issue lifecycle: search (with nextPageToken paging), CRUD, assign, comments, transitions (with an optional comment), attachments, issue links, worklogs, watchers, subtasks, labels, history, agile boards, sprints (incl. moving issues into a sprint), versions, batch creation (up to 50 issues), dev status, epic linking, and user lookup.

• Bitbucket (30 tools) -- Repository management, branches, commits, pull requests (create/update/list, reviewers, approve/request changes, merge/decline), PR comments with resolve/unresolve, PR diffstat and per-file diffs, file content at a ref, CI/pipeline status and step logs, and issue tracking.

• Smart Field Handling -- AI-driven field suggestions during Jira transitions. Analyzes issue context to auto-suggest values for required fields (e.g., DB scripts, test cases), reducing manual input.

• Snapshot Safety Net -- Automatic local snapshots of mutating operations (create/update/delete) with configurable retention, so you can recover from accidental changes.

Related MCP server: Jira MCP Server

Prerequisites

• Node.js >= 18

• Atlassian Cloud account(s) with API tokens

• An MCP-compatible client (e.g., Claude Desktop)

Installation

From npm (recommended)

npm install -g @jagadeesh52423/atlassian-mcp-server

Or run without installing:

npx @jagadeesh52423/atlassian-mcp-server

From source

git clone https://github.com/jagadeeshpulamarasetti/atlassian-mcp-server.git
cd atlassian-mcp-server
npm install
npm run build

Configuration

1. Set environment variables

Copy the template and fill in your credentials:

cp .env.example .env

# Confluence
CONFLUENCE_DOMAIN=your-domain.atlassian.net
CONFLUENCE_EMAIL=your-email@example.com
CONFLUENCE_API_TOKEN=your-confluence-api-token

# Jira
JIRA_DOMAIN=your-domain.atlassian.net
JIRA_EMAIL=your-email@example.com
JIRA_API_TOKEN=your-jira-api-token

# Bitbucket
BITBUCKET_WORKSPACE=your-workspace
BITBUCKET_USERNAME=your-username
BITBUCKET_API_TOKEN=your-bitbucket-api-token

You can also set legacy ATLASSIAN_* variables as a fallback for Confluence and Jira. Bitbucket always requires its own credentials.

# Jira OAuth (advanced auth)
JIRA_CLIENT_ID=your-client-id
JIRA_CLIENT_SECRET=your-client-secret

# Snapshot safety net (defaults shown)
MCP_SNAPSHOTS_ENABLED=true
MCP_SNAPSHOT_DIR=~/.atlassian-mcp-snapshots
MCP_SNAPSHOT_RETENTION_DAYS=30

2. Get an API token

1. Go to Atlassian API Tokens

2. Click Create API token, give it a name, and copy the value

3. For Bitbucket, create a token under Personal settings > Access tokens with repo, issue, and PR scopes

Usage

Running the server

# Global install
atlassian-mcp-server

# From source
npm start

# Development (auto-reload)
npm run dev

Claude Desktop integration

Add one of the following to your Claude Desktop config:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "atlassian": {
      "command": "npx",
      "args": ["-y", "@jagadeesh52423/atlassian-mcp-server"],
      "env": {
        "CONFLUENCE_DOMAIN": "your-domain.atlassian.net",
        "CONFLUENCE_EMAIL": "your-email@example.com",
        "CONFLUENCE_API_TOKEN": "your-token",
        "JIRA_DOMAIN": "your-domain.atlassian.net",
        "JIRA_EMAIL": "your-email@example.com",
        "JIRA_API_TOKEN": "your-token",
        "BITBUCKET_WORKSPACE": "your-workspace",
        "BITBUCKET_USERNAME": "your-username",
        "BITBUCKET_API_TOKEN": "your-token"
      }
    }
  }
}

Global install:

{
  "mcpServers": {
    "atlassian": {
      "command": "atlassian-mcp-server",
      "env": { "..." : "same env vars as above" }
    }
  }
}

From source:

{
  "mcpServers": {
    "atlassian": {
      "command": "node",
      "args": ["/path/to/atlassian-mcp-server/dist/index.js"]
    }
  }
}

Available Tools

Confluence (22 tools)

Jira (50 tools)

All Jira tools are prefixed with jira_.

Bitbucket (30 tools)

PR comment resolution (v1.3.0+)

• bitbucket_resolve_pr_comment / bitbucket_unresolve_pr_comment use Bitbucket's dedicated resolution sub-resource (POST/DELETE .../comments/{id}/resolve) and return the refreshed comment.

• Top-level comments only -- Bitbucket only allows resolving/unresolving top-level review comments, not replies. Calling these on a reply returns Comment is not a top-level comment.

• Comments returned by bitbucket_get_pr_comments, bitbucket_resolve_pr_comment, and bitbucket_unresolve_pr_comment now include a resolved boolean. When a comment is resolved, the single-comment responses also include resolvedOn (ISO timestamp) and resolvedBy (display name); these resolution details are not present in the bulk list response.

Confluence Content Format (v1.2.0+)

Confluence pages and comments use ADF (Atlassian Document Format) natively:

• Input -- create_page, update_page, add_comment, update_comment accept Markdown, which the server converts to ADF using the official @atlaskit/editor-markdown-transformer.

• Output -- get_page, get_comments return ADF JSON (an object, not a string).

• Comments -- get_comments returns both footer (page-level) and inline (text-anchored) comments by default. Use the commentType param (footer | inline | all, default all) to filter. Every comment carries a type field ("footer" or "inline"). Inline comments additionally include anchoredText (the highlighted page text the comment is attached to), resolutionStatus (e.g. "open", "resolved"), markerRef, and webuiLink. Backwards-compatibility note: callers that previously relied on get_comments returning footer comments only will now also receive inline comments under the default all; pass commentType: "footer" to restore the old footer-only behaviour. The new type field is additive and does not remove any existing field.

• Inline comment anchors -- An inline comment is not a separate resource: it anchors to page text via an annotation mark inside the page body. Any update that rebuilds the body therefore detaches every inline comment on the page. update_page guards against this by reading the live body first and re-anchoring each marker onto text that survived the edit, then reporting inlineComments: { preserved, dropped, ambiguous }. A comment whose text you reworded or deleted appears in dropped; one whose text now occurs in more than one place appears in ambiguous, because the server refuses to pick an occurrence — a wrong guess silently moves a reviewer's question onto unrelated text while reporting success. Either way the comment stays on the page but unattached, which is exactly what Confluence does today. Prefer leaving commented sentences intact. To repair comments detached by an earlier update (including ones made before this feature existed), use confluence_reattach_inline_comments: it walks page history to find where each orphan was anchored, re-marks only unambiguous matches, and reports the rest rather than guessing. It never alters page text, and searches back at most 50 versions. Anchors are keyed by surrounding content (table header + first-column value, or block type + text) rather than by position, so they survive rows and paragraphs being inserted above them — which is what lets a comment on a repeated field name like unique_id resolve even though that text appears ten times on the page. Note that anchoredText from get_comments is truncated by Confluence at ~250 characters (252 observed); the history lookup recovers the full anchor.

• Inline images -- Not currently supported via ADF media nodes (requires Atlassian Media API tokens). Use confluence_add_attachment instead.

Project Structure

src/
├── index.ts              # Server entry point and tool routing
├── config.ts             # Environment variable loading and validation
├── types.ts              # TypeScript interfaces
├── constants.ts          # Shared constants
├── error-handler.ts      # Structured error responses
├── formatters/           # Markdown-to-ADF and ADF-to-text converters
├── clients/              # HTTP clients (one per Atlassian product)
├── services/             # Business logic for each product
│   └── jira/             # Jira sub-modules (agile, comments, transitions, etc.)
├── handlers/             # MCP request handlers
├── tools/                # MCP tool definitions (schemas)
├── registry/             # Tool dispatch registry
└── snapshot/             # Snapshot safety net (auto-backup before mutations)

Security

• Credentials are read from environment variables only; never committed to source control.

• All API calls use HTTPS with Basic Auth (base64-encoded email:token).

• Grant tokens the minimum required permissions.

• Snapshots are stored locally in ~/.atlassian-mcp-snapshots by default.

License

MIT

Links

• npm: @jagadeesh52423/atlassian-mcp-server

• Source: GitHub

• Issues: GitHub Issues