Confluence Knowledge Base MCP Server

An MCP server that turns your Confluence documentation into an AI-powered knowledge base for Gemini CLI. Ask natural language questions about your systems and get answers from your actual documentation.

Quick Start

One-Command Setup

The interactive wizard will:

1. ✅ Install dependencies in a virtual environment

2. ✅ Ask for your Confluence credentials

3. ✅ Discover your spaces

4. ✅ Help you choose which spaces to index

5. ✅ Build the initial knowledge base

6. ✅ Configure Gemini CLI automatically (merges with existing config)

What You'll Need

Before running the installer:

1. Confluence API Token

   • Go to: https://id.atlassian.com/manage-profile/security/api-tokens

   • Click "Create API token"

   • Copy the token (you won't see it again!)

2. Your Confluence URL

   • Example: https://yourcompany.atlassian.net

3. Python 3.8+ installed

   • The installer creates a virtual environment automatically (no system-wide packages needed)

4. Gemini CLI installed

   • Install from: https://github.com/google-gemini/gemini-cli

Usage

Once installed, just start Gemini CLI and ask questions:

Gemini will automatically retrieve relevant documentation and answer your questions!

How It Works

Technologies Used

• FastMCP - MCP server framework

• ChromaDB - Local vector database

• sentence-transformers - Semantic search

• Confluence REST API - Documentation retrieval

Project Structure

Configuration

After installation, configuration is stored in:

• Credentials: ~/.confluence_mcp.env

• Index: ~/.confluence_mcp/index/

• Gemini Config: ~/.gemini/settings.json

• Virtual Environment: ./venv/ (in the project directory)

Updating Documentation

When your Confluence docs are updated:

Option 1: Ask Gemini

Option 2: Command line

Option 3: Automated (Weekly)

Set up a cron job (see REINDEXING_GUIDE.md)

Customization

Change indexed spaces

Edit ~/.confluence_mcp.env:

Then rebuild the index.

Adjust chunk size

In confluence_knowledge_base.py:

Change embedding model

For better quality (slower, larger):

Troubleshooting

"Connection failed"

Check that:

• Your Confluence URL is correct

• Your API token is valid

• You have internet connectivity

"No spaces found"

You might not have access to any Confluence spaces. Ask your admin for access.

Slow indexing

Normal for large documentation sets (500+ pages). Reduce spaces or run overnight.

Wrong/outdated answers

Your index is cached! Reindex when docs are updated:

Advanced Usage

Manual space discovery

Staleness detection

Use the enhanced version with automatic staleness warnings:

Add environment variables:

Scheduled reindexing

See REINDEXING_GUIDE.md for cron job setup.

FAQ

Q: Does this modify my Confluence documentation? A: No, it's read-only. It only downloads and indexes content.

Q: Where is my data stored? A: Locally in ~/.confluence_mcp/index/. Nothing is sent to external services except Gemini API calls.

Q: How much does it cost? A: The MCP server is free. You only pay for Gemini API usage (queries to the AI).

Q: Can I use this with Claude instead of Gemini? A: Yes! MCP is a standard protocol. Just configure Claude Desktop to use this MCP server.

Q: How often should I reindex? A: Depends on how often your docs are updated. Weekly is common. Daily if very active.

Q: Can I exclude certain pages? A: Not by default, but you can modify confluence_knowledge_base.py to filter by title, label, etc.

Q: What about attachments/PDFs? A: Currently only page content is indexed. Attachments could be added with additional code.

Documentation

• KNOWLEDGE_BASE_SETUP.md - Comprehensive setup guide

• REINDEXING_GUIDE.md - Strategies for keeping docs fresh

Contributing

Feel free to:

• Add features (write capabilities, attachment support, etc.)

• Improve chunking strategies

• Add better error handling

• Create additional tools

License

[Your license here]

Support

For issues or questions:

1. Check the troubleshooting section above

2. Review the detailed guides in /docs

3. Open an issue on GitHub

Ready to get started? Just run: