Confluence知识库MCP服务器

Atlassian Confluence MCP Server

An open-source Model Context Protocol (MCP) server for Atlassian Confluence that lets AI assistants like Claude and Windsurf read, create, search, and manage Confluence wiki pages. Works with Confluence Cloud, on-premise Server, and Data Center deployments — connect your AI coding agent to your team's knowledge base in seconds via npx.

Why atlassian-confluence-mcp-server?

• ✅ Only MCP server supporting Cloud + Server + Data Center in one package
• ✅ 47 tools — most comprehensive Confluence MCP available
• ✅ Zero config via npx — no install needed
• ✅ PAT, Basic Auth, and API token support
• ✅ Actively maintained with provenance-signed releases

Why use this instead of Atlassian's official MCP server?

Features

• Space Management — Create, delete, list, and inspect spaces
• Page CRUD — Create, read, update, delete, copy, and move pages
• Search — Full CQL and text search, recently modified pages
• Labels — Add, list, and remove page labels
• Comments — Read and post page comments
• Attachments — List and upload page and comment attachments
• Page Hierarchy — Navigate parent/child relationships
• Version History — Browse and inspect page versions
• Watchers — View and manage page watchers
• Permissions — Get and set page restrictions
• Tasks — Extract inline tasks from pages
• Export — Generate PDF/Word export URLs
• Personal Space — Create pages in your personal space
• URL Parsing — Retrieve page content from a Confluence URL

Prerequisites

• Node.js 22.12+
• A Confluence instance (Cloud or Server/Data Center 7.9+)
• Authentication credentials (PAT or username/password)

Installation

Via npx (no install required)

npx atlassian-confluence-mcp-server

Global install

npm install -g atlassian-confluence-mcp-server

From source

git clone https://github.com/anuragagrawal0430/atlassian-confluence-mcp-server.git
cd atlassian-confluence-mcp-server
npm install
npm run build

Configuration

All configuration is passed through environment variables. Never hard-code credentials.

Confluence Cloud

Cloud API tokens are generated from your Atlassian account and require both your email and the token:

CONFLUENCE_BASE_URL=https://your-site.atlassian.net
CONFLUENCE_USERNAME=you@example.com
PAT=your-api-token

1. Go to https://id.atlassian.com/manage-profile/security/api-tokens
2. Click Create API token, give it a label, and copy the value
3. Set CONFLUENCE_USERNAME to your Atlassian account email
4. Set PAT to the copied API token

Confluence Server / Data Center (7.9+)

Server and Data Center instances use a Personal Access Token with Bearer authentication. Only the token is needed:

CONFLUENCE_BASE_URL=https://confluence.example.com
PAT=your-personal-access-token

1. Navigate to Profile > Settings > Personal Access Tokens
2. Create a token with the required permissions
3. Copy the token value

Basic Authentication (fallback)

For older Server versions without PAT support, use username/password:

CONFLUENCE_BASE_URL=https://confluence.example.com
CONFLUENCE_USERNAME=your-username
CONFLUENCE_PASSWORD=your-password

Security Controls (important)

The server now uses secure defaults to reduce prompt-injection blast radius and credential leakage risk:

• CONFLUENCE_READ_ONLY=true by default
  • Mutating tools are disabled unless you set CONFLUENCE_READ_ONLY=false.
• CONFLUENCE_ENABLE_DESTRUCTIVE_TOOLS=false by default
  • confluence_delete_page, confluence_delete_space, and confluence_set_page_permissions remain disabled unless explicitly enabled.
• CONFLUENCE_ENABLED_TOOLS (optional)
  • Restrict exposure to an explicit tool allowlist.
• CONFLUENCE_ALLOWED_SPACES (optional)
  • Restrict requests that include spaceKey / destinationSpaceKey to an approved set.
• CONFLUENCE_ALLOW_INSECURE_HTTP=false by default
  • HTTPS is required unless this is set to true.
  • Even when enabled, insecure HTTP is restricted to localhost/loopback only.
• CONFLUENCE_MAX_ATTACHMENT_BYTES (optional, default 50 MB)
  • Caps how large a host file confluence_upload_attachment will read before calling Confluence.
  • Independent of the instance "Attachment Maximum Size" admin setting (not exposed via public REST).
• CONFLUENCE_UPLOAD_ALLOWED_DIRS (optional, defaults to empty)
  • Explicitly restricts which host directories the MCP server is allowed to read files from when uploading attachments.
  • If not configured, file uploads from the host filesystem are entirely disabled.

Example (enable safe writes, keep destructive tools disabled):

CONFLUENCE_READ_ONLY=false
CONFLUENCE_ENABLE_DESTRUCTIVE_TOOLS=false

Example (strict allowlist):

CONFLUENCE_ENABLED_TOOLS=confluence_get_page,confluence_search_pages,confluence_get_page_body_chunk

Example (space-level allowlist):

CONFLUENCE_ALLOWED_SPACES=ENG,SECURITY

MCP Client Configuration

Cloud (npx)

{
  "mcpServers": {
    "confluence": {
      "command": "npx",
      "args": ["-y", "atlassian-confluence-mcp-server"],
      "env": {
        "CONFLUENCE_BASE_URL": "https://your-site.atlassian.net",
        "CONFLUENCE_USERNAME": "you@example.com",
        "PAT": "your-api-token"
      }
    }
  }
}

Server / Data Center (npx)

{
  "mcpServers": {
    "confluence": {
      "command": "npx",
      "args": ["-y", "atlassian-confluence-mcp-server"],
      "env": {
        "CONFLUENCE_BASE_URL": "https://confluence.example.com",
        "PAT": "your-personal-access-token"
      }
    }
  }
}

Global install

{
  "mcpServers": {
    "confluence": {
      "command": "atlassian-confluence-mcp-server",
      "env": {
        "CONFLUENCE_BASE_URL": "https://confluence.example.com",
        "PAT": "your-personal-access-token"
      }
    }
  }
}

From source

{
  "mcpServers": {
    "confluence": {
      "command": "node",
      "args": ["./dist/index.js"],
      "env": {
        "CONFLUENCE_BASE_URL": "https://confluence.example.com",
        "PAT": "your-personal-access-token"
      }
    }
  }
}

Available Tools (47)

Note: tool exposure depends on security configuration. With the default CONFLUENCE_READ_ONLY=true, mutating tools are intentionally hidden/blocked.

Connection

Spaces

Pages

Search

Labels

Comments

Attachments

Page Hierarchy

Version History

Properties

User

Watchers

Permissions

Tasks

Export

Utilities

Personal Space

Resources

Example CQL Queries

# Pages in a specific space
type=page AND space=TEAM

# Pages with a label
type=page AND label=documentation

# Recently modified pages
type=page AND lastModified > now("-7d")

# Title search
type=page AND title~"Meeting Notes"

# Pages created by current user
type=page AND creator=currentUser()

Page Body Format

Pages use Confluence storage format (XHTML):

<p>This is a paragraph.</p>
<h1>Heading 1</h1>
<ul>
  <li>Item 1</li>
  <li>Item 2</li>
</ul>
<ac:structured-macro ac:name="code">
  <ac:parameter ac:name="language">javascript</ac:parameter>
  <ac:plain-text-body><![CDATA[console.log("Hello");]]></ac:plain-text-body>
</ac:structured-macro>

Editing Very Large Pages

When a page body is very large (for example hundreds of KB), avoid sending the full body through confluence_update_page.

Recommended workflow:

1. Use confluence_get_page_body_chunk to inspect the relevant part of the page.
2. Use one of:
   • confluence_patch_page for exact find/replace edits.
   • confluence_replace_page_range for replacing a specific middle range.
   • confluence_append_to_page for incremental append/prepend updates.

Notes:

• Offset-based range replacement is the most reliable mode for storage XHTML.
• Line-based replacement requires actual newline characters in the stored body.
• These tools still perform a full Confluence update under the hood, but the full body remains server-side.

Security

• Credentials are read exclusively from environment variables — never hard-coded.
• Input validation — all user-supplied parameters are validated and sanitized before use.
• CQL injection prevention — user input interpolated into CQL queries is escaped.
• URL path encoding — all dynamic path segments are encoded with encodeURIComponent.
• Request timeouts — all HTTP requests have a 30-second timeout to prevent hanging.
• Error sanitization — API error responses are truncated and stripped of HTML before surfacing.
• Pagination limits — query limits are clamped to a safe maximum (100) to prevent abuse.

Contributing

Please see CONTRIBUTING.md for guidelines on how to contribute to this project.

Security Policy

For security policies and vulnerability reporting, see SECURITY.md.

Changelog

See CHANGELOG.md for a history of changes.

License

MIT

Acknowledgments

• Model Context Protocol
• Atlassian Confluence REST API