Connect AI to Your Confluence Knowledge Base

Transform how you access and interact with your team's knowledge by connecting Claude, Cursor AI, and other AI assistants directly to your Confluence spaces, pages, and documentation. Get instant answers from your knowledge base, search across all your spaces, and streamline your documentation workflow.

What You Can Do

• Ask AI about your documentation: "What's our API authentication process?"

• Search across all spaces: "Find all pages about security best practices"

• Get instant answers: "Show me the latest release notes from the Product space"

• Access team knowledge: "What are our HR policies for remote work?"

• Review page comments: "Show me the discussion on the architecture document"

• Create and update content: "Create a new page in the DEV space"

Related MCP server: Confluence MCP

Perfect For

• Developers who need quick access to technical documentation and API guides

• Product Managers searching for requirements, specs, and project updates

• HR Teams accessing policy documents and employee resources quickly

• Support Teams finding troubleshooting guides and knowledge base articles

• Anyone who wants to interact with Confluence using natural language

Quick Start

Get up and running in 2 minutes:

1. Get Your Confluence Credentials

Generate a Confluence API Token:

1. Go to Atlassian API Tokens

2. Click Create API token

3. Give it a name like "AI Assistant"

4. Copy the generated token immediately (you won't see it again!)

2. Try It Instantly

Connect to AI Assistants

For Claude Desktop Users

Add this to your Claude configuration file (~/.claude/claude_desktop_config.json):

Restart Claude Desktop, and you'll see the confluence server in the status bar.

For Other AI Assistants

Most AI assistants support MCP (Cursor AI, Continue.dev, and others). Install the server globally:

Then configure your AI assistant to use the MCP server with STDIO transport. The binary is available as mcp-atlassian-confluence after global installation.

Alternative: Configuration File

Create ~/.mcp/configs.json for system-wide configuration:

Alternative config keys: The system also accepts "atlassian-confluence", "@aashari/mcp-server-atlassian-confluence", or "mcp-server-atlassian-confluence" instead of "confluence".

Using Environment Variables

You can also configure credentials using environment variables or a .env file:

The server will automatically load these values from:

1. Environment variables

2. .env file in the current directory

3. ~/.mcp/configs.json (as shown above)

Available Tools

This MCP server provides 5 generic tools that can access any Confluence API endpoint:

Tool Parameters

All tools share these common parameters:

• path (required): The API endpoint path (e.g., /wiki/api/v2/spaces)

• queryParams (optional): Query parameters as key-value pairs (e.g., {"limit": "25", "space-id": "123"})

• jq (optional): JMESPath expression to filter/transform the response (e.g., results[*].{id: id, title: title})

• outputFormat (optional): Output format - "toon" (default, 30-60% fewer tokens) or "json"

Tools that accept a request body (conf_post, conf_put, conf_patch):

• body (required): Request body as a JSON object

Common API Paths

Spaces:

• /wiki/api/v2/spaces - List all spaces

• /wiki/api/v2/spaces/{id} - Get space details

Pages:

• /wiki/api/v2/pages - List pages (use space-id query param to filter)

• /wiki/api/v2/pages/{id} - Get page details

• /wiki/api/v2/pages/{id}/body - Get page body (use body-format param)

• /wiki/api/v2/pages/{id}/children - Get child pages

• /wiki/api/v2/pages/{id}/labels - Get page labels

Comments:

• /wiki/api/v2/pages/{id}/footer-comments - List/add footer comments

• /wiki/api/v2/pages/{id}/inline-comments - List/add inline comments

• /wiki/api/v2/footer-comments/{comment-id} - Get/update/delete comment

Blog Posts:

• /wiki/api/v2/blogposts - List blog posts

• /wiki/api/v2/blogposts/{id} - Get blog post

Search:

• /wiki/rest/api/search - Search content (use cql query param)

TOON Output Format

What is TOON? TOON (Token-Oriented Object Notation) is a format optimized for LLM token efficiency, reducing token costs by 30-60% compared to JSON. It's the default output format for all tools.

Benefits:

• Tabular arrays use fewer tokens than JSON arrays

• Minimal syntax overhead (no quotes, brackets, commas where unnecessary)

• Still human-readable and parseable

When to use JSON instead:

• When you need standard JSON for other tools

• When debugging or manual inspection is needed

Example comparison:

To use JSON instead of TOON, set outputFormat: "json" in your request.

JMESPath Filtering

All tools support optional JMESPath (jq) filtering to extract specific data and reduce token costs:

IMPORTANT: Always use the jq parameter to filter responses to only the fields you need. Unfiltered responses can be very large and expensive in token costs.

JMESPath Syntax Reference:

• Official docs: jmespath.org

• Common patterns:

  • results[*] - All items in results array

  • results[0] - First item only

  • results[*].id - Just IDs from all items

  • results[*].{id: id, title: title} - Create objects with selected fields

  • results[?status=='current'] - Filter by condition

Real-World Examples

Explore Your Knowledge Base

Ask your AI assistant:

• "List all the spaces in our Confluence"

• "Show me details about the Engineering space"

• "What pages are in our Product space?"

• "Find the latest pages in the Marketing space"

Search and Find Information

Ask your AI assistant:

• "Search for pages about API authentication"

• "Find all documentation with 'security' in the title"

• "Show me pages labeled with 'getting-started'"

• "Search for content in the DEV space about deployment"

Access Specific Content

Ask your AI assistant:

• "Get the content of the API Authentication Guide page"

• "Show me the onboarding checklist document"

• "What's in our security policies page?"

• "Display the latest release notes"

Create and Update Content

Ask your AI assistant:

• "Create a new page in the DEV space titled 'API Guide'"

• "Add a comment to the architecture document"

• "Update the page content with the new release info"

CLI Commands

The CLI mirrors the MCP tools for direct terminal access. All commands support the same parameters as the tools.

Available Commands

• get - GET any Confluence endpoint

• post - POST to any endpoint

• put - PUT to any endpoint

• patch - PATCH any endpoint

• delete - DELETE from any endpoint

CLI Parameters

All commands:

• -p, --path <path> (required) - API endpoint path

• -q, --query-params <json> (optional) - Query parameters as JSON

• --jq <expression> (optional) - JMESPath filter expression

• -o, --output-format <format> (optional) - Output format: toon (default) or json

Commands with body (post, put, patch):

• -b, --body <json> (required) - Request body as JSON

Examples

Response Handling

Large Response Truncation

When API responses exceed approximately 40,000 characters (~10,000 tokens), the server automatically truncates the response to stay within token limits. When this happens:

1. You'll see a truncation notice at the end of the response showing:

   • How much of the original response is shown

   • The original response size

   • Guidance on accessing the full data

2. The full raw response is saved to a temporary file in /tmp/mcp/ (path provided in the truncation notice)

3. Best practices to avoid truncation:

   • Always use the to filter responses to only needed fields

   • Use limit query parameter to restrict result counts (e.g., {"limit": "5"})

   • Request specific resources by ID rather than listing all

   • Use targeted CQL queries for searches

Example of efficient filtering:

Debug Logging

Enable debug logging to see detailed request/response information:

Debug logs are written to: ~/.mcp/data/@aashari-mcp-server-atlassian-confluence.[session-id].log

Testing & Development

Using MCP Inspector

The MCP Inspector provides a visual interface for testing tools:

Or use the built-in development command if you've cloned the repository:

This starts the server in HTTP mode and opens the inspector UI in your browser.

HTTP Mode for Testing

You can run the server in HTTP mode to test with curl or other HTTP clients:

The server will listen on http://localhost:3000/mcp by default. You can change the port:

Testing with curl:

The response comes as Server-Sent Events (SSE) with format:

Troubleshooting

"Authentication failed" or "403 Forbidden"

1. Check your API Token permissions:

   • Go to Atlassian API Tokens

   • Make sure your token is still active and hasn't expired

2. Verify your site name format:

   • If your Confluence URL is https://mycompany.atlassian.net

   • Your site name should be just mycompany

3. Test your credentials:

   npx -y @aashari/mcp-server-atlassian-confluence get --path "/wiki/api/v2/spaces?limit=1"

"Resource not found" or "404"

1. Check the API path:

   • Paths are case-sensitive

   • Use numeric IDs for spaces and pages (not keys)

   • Verify the resource exists in your browser

2. Verify access permissions:

   • Make sure you have access to the space/page in your browser

   • Some content may be restricted to certain users

"No results found" when searching

1. Try different search terms:

   • Use CQL syntax for advanced searches

   • Try broader search criteria

2. Check CQL syntax:

   • Validate your CQL in Confluence's advanced search first

Claude Desktop Integration Issues

1. Restart Claude Desktop after updating the config file

2. Verify config file location:

   • macOS: ~/.claude/claude_desktop_config.json

   • Windows: %APPDATA%\Claude\claude_desktop_config.json

Getting Help

If you're still having issues:

1. Run a simple test command to verify everything works

2. Check the GitHub Issues for similar problems

3. Create a new issue with your error message and setup details

Frequently Asked Questions

What permissions do I need?

Your Atlassian account needs:

• Access to Confluence with the appropriate permissions for the spaces you want to query

• API token with appropriate permissions (automatically granted when you create one)

Can I use this with Confluence Server (on-premise)?

Currently, this tool only supports Confluence Cloud. Confluence Server/Data Center support may be added in future versions.

How do I find my site name?

Your site name is the first part of your Confluence URL:

• URL: https://mycompany.atlassian.net -> Site name: mycompany

• URL: https://acme-corp.atlassian.net -> Site name: acme-corp

What AI assistants does this work with?

Any AI assistant that supports the Model Context Protocol (MCP):

• Claude Desktop

• Cursor AI

• Continue.dev

• Many others

Is my data secure?

Yes! This tool:

• Runs entirely on your local machine

• Uses your own Confluence credentials

• Never sends your data to third parties

• Only accesses what you give it permission to access

Can I search across all my spaces at once?

Yes! Use CQL queries for cross-space searches. For example:

Migration from v2.x

Version 3.0 replaces 8+ specific tools with 5 generic HTTP method tools. If you're upgrading from v2.x:

Before (v2.x):

After (v3.0):

Migration examples:

• conf_ls_spaces -> conf_get with path /wiki/api/v2/spaces

• conf_get_space -> conf_get with path /wiki/api/v2/spaces/{id}

• conf_ls_pages -> conf_get with path /wiki/api/v2/pages?space-id={id}

• conf_get_page -> conf_get with path /wiki/api/v2/pages/{id}

• conf_search -> conf_get with path /wiki/rest/api/search?cql=...

• conf_add_comment -> conf_post with path /wiki/api/v2/pages/{id}/footer-comments

Technical Details

Requirements

• Node.js: 18.0.0 or higher

• MCP SDK: 1.23.0 (uses modern registerTool API)

• Confluence: Cloud only (Server/Data Center not supported)

Architecture

This server follows a 5-layer architecture:

1. Tools Layer (src/tools/) - MCP tool definitions with Zod validation

2. CLI Layer (src/cli/) - Commander-based CLI for direct testing

3. Controllers Layer (src/controllers/) - Business logic, JMESPath filtering, output formatting

4. Services Layer (src/services/) - Confluence API communication

5. Utils Layer (src/utils/) - Shared utilities (logger, config, formatters, TOON encoder)

Features

• Generic HTTP method tools - Access any Confluence API endpoint

• TOON output format - 30-60% token reduction vs JSON

• JMESPath filtering - Extract only needed data

• Response truncation - Automatic handling of large responses

• Raw response logging - Full responses saved to /tmp/mcp/

• Dual transport - STDIO (for Claude Desktop) and HTTP (for web integrations)

• Debug logging - Comprehensive logging for troubleshooting

Version History

v3.2.1 (Current)

• Add raw response logging with truncation for large API responses

• Improve dependency compatibility

v3.2.0

• Modernize MCP SDK to v1.23.0 with registerTool API

v3.1.0

• Add TOON output format for token-efficient LLM responses

v3.0.0 (Breaking change)

• Replace 8+ domain-specific tools with 5 generic HTTP method tools

• Add JMESPath filtering support

• Full Confluence API access via generic methods

See CHANGELOG.md for complete version history.

Support

Need help? Here's how to get assistance:

1. Check the troubleshooting section above - most common issues are covered there

2. Visit our GitHub repository for documentation and examples: github.com/aashari/mcp-server-atlassian-confluence

3. Report issues at GitHub Issues

4. Start a discussion for feature requests or general questions

Made with care for teams who want to bring AI into their knowledge management workflow.