S3 Documentation MCP Server

- 📚 Product Documentation: Let Claude/Cursor/etc answer from your docs
- 🏢 Internal Wiki: AI-powered company knowledge search
- 📖 API Docs: Help developers find API information
- 🎓 Educational Content: Build AI tutors with course materials

# 1. Prerequisites # Install Ollama from https://ollama.ai ollama pull nomic-embed-text # 2. Configure cp env.example .env # Add your S3 credentials # 3. Run docker run -d \ --name s3-doc-mcp \ -p 3000:3000 \ --env-file .env \ -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \ -v $(pwd)/data:/app/data \ yoanbernabeu/s3-doc-mcp:latest

# 1. Prerequisites # Install Ollama from https://ollama.ai ollama pull nomic-embed-text # 2. Install & Run npm install cp env.example .env # Configure your S3 credentials npm run build && npm start # 3. For local development npm run dev

Your MCP server is now running onhttp://localhost:3000

Once your server is running, you need to configure your MCP client to connect to it.

Edit your~/.cursor/mcp.jsonfile and add:

{ "mcpServers": { "doc": { "type": "streamable-http", "url": "http://127.0.0.1:3000/mcp", "note": "S3 Documentation RAG Server" } } }

Edit your Claude Desktop configuration file:

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

{ "mcpServers": { "doc": { "type": "streamable-http", "url": "http://127.0.0.1:3000/mcp", "note": "S3 Documentation RAG Server" } } }

Restart your MCP client, and you should now see:

- 3 MCP Tools:search_documentation,refresh_index,get_full_document
- MCP Resources: Full list of indexed documentation files with direct access

💡Tip: If using Docker, make sure the port mapping matches your configuration (default is3000:3000)

- 🔌Universal S3: AWS S3, MinIO, Scaleway, DigitalOcean Spaces, Cloudflare R2, Wasabi...
- 🧠Flexible Embeddings:

- Splits documents into chunks (1000 characters by default)
- Generates embeddings using your chosen provider:

- Tools:search_documentation,refresh_index,get_full_documentfor semantic search and actions
- Resources:resources/list,resources/readfor file discovery and direct access

- ⚡Fast: Approximate nearest neighbor search in milliseconds
- 💾Simple: Stores indices as local files (no database needed)
- 🪶Efficient: Low memory footprint, ideal for personal/small-team documentation
- 🎯Accurate: High recall with cosine similarity for semantic search

It's the sweet spot between simplicity and performance for RAG applications.

Copyenv.exampleto.envand configure your environment variables:

# S3 Configuration S3_BUCKET_NAME=your-bucket-name # Your S3 bucket name S3_ACCESS_KEY_ID=your-access-key # S3 access key S3_SECRET_ACCESS_KEY=your-secret-key # S3 secret key S3_REGION=us-east-1 # S3 region S3_ENDPOINT= # Optional: for non-AWS S3 (MinIO, Scaleway, etc.) # Embeddings Provider (choose one) EMBEDDING_PROVIDER=ollama # ollama (default) or openai # Option 1: Ollama (Local) OLLAMA_BASE_URL=http://localhost:11434 # Ollama API endpoint OLLAMA_EMBEDDING_MODEL=nomic-embed-text # Ollama embedding model # Option 2: OpenAI (Cloud) - Only if EMBEDDING_PROVIDER=openai OPENAI_API_KEY= # Your OpenAI API key OPENAI_EMBEDDING_MODEL=text-embedding-3-small # or text-embedding-3-large

The server supports two embedding providers:

- ✅Free: No API costs, unlimited usage
- ✅Private: All data stays on your machine
- ✅Offline: Works without internet connection
- ✅Fast: Direct local API calls

- ⚠️ Requires Ollama installation and model download
- ⚠️ Uses local CPU/GPU resources

# Install Ollama from https://ollama.ai ollama pull nomic-embed-text # Configure EMBEDDING_PROVIDER=ollama OLLAMA_BASE_URL=http://localhost:11434 OLLAMA_EMBEDDING_MODEL=nomic-embed-text

- ✅High Accuracy: State-of-the-art embeddings
- ✅Multilingual: Excellent support for 20+ languages
- ✅No Local Resources: Runs entirely in the cloud
- ✅Lower Latency: Fast API responses

- ⚠️ Requires API key and credits
- ⚠️ Data sent to OpenAI servers
- ⚠️ Cost per token (very affordable: ~$0.00002/1K tokens fortext-embedding-3-small)

# Get an API key from https://platform.openai.com/api-keys # Configure EMBEDDING_PROVIDER=openai OPENAI_API_KEY=sk-...your-key... OPENAI_EMBEDDING_MODEL=text-embedding-3-small # or text-embedding-3-large

💡Tip: Start withtext-embedding-3-smallfor most use cases. Only switch totext-embedding-3-largeif you need the absolute best accuracy or work extensively with non-English content.

If you setEMBEDDING_PROVIDER=openaibut don't provide a validOPENAI_API_KEY, the server will automatically fall back to Ollama (if configured). This ensures the server can always start, even with incomplete configuration.

The server supports three synchronization modes viaSYNC_MODE:

-

startup(default): Syncs at server startup

- ✅Auto-detection: If the vector store is empty, automatically performs a full sync
- ✅ Otherwise, performs an incremental sync (only changed files)
- ✅ No manualrefresh_indexneeded after restart!

periodic: Syncs at regular intervals (SYNC_INTERVAL_MINUTES)

- You must callrefresh_indextool manually

💡Note: The server automatically detects when the vector store is empty (e.g., after deleting./data/folder or first run) and triggers a full synchronization. You no longer need to manually runrefresh_indexafter every restart!

By default, the server runs inopen access modefor easy local development. For shared or remote deployments, you can enable API key authentication:

# Enable authentication ENABLE_AUTH=true # Set your API key MCP_API_KEY=your-secret-key-here

- ✅ All endpoints (except/health) require a valid API key
- ✅ API key can be provided via:

- Authorization header(recommended):Authorization: Bearer your-secret-key
- Query parameter:?api_key=your-secret-key

# With Authorization header (recommended) curl -H "Authorization: Bearer your-secret-key" http://localhost:3000/mcp # With query parameter curl "http://localhost:3000/mcp?api_key=your-secret-key"

{ "mcpServers": { "doc": { "type": "streamable-http", "url": "http://127.0.0.1:3000/mcp", "headers": { "Authorization": "Bearer your-secret-key" }, "note": "S3 Documentation RAG Server with authentication" } } }

- Keep authenticationdisabledfor local development
- Enableit for shared networks or remote deployments
- Use strong, randomly generated keys (e.g.,openssl rand -hex 32)
- The/healthendpoint is always accessible without authentication for monitoring

{ "query": "How to configure S3?", "max_results": 4 }

Returns relevant document chunks with similarity scores and sources.

{ "force": false // default: incremental sync (recommended) }

Synchronizes the documentation index with S3, detecting new, modified, or deleted files.

- force(boolean, optional, default:false)

- false:Incremental sync- Only processes changes (fast, efficient) ✅
- true:Full reindex- Reprocesses ALL files (slow, expensive) ⚠️

⚠️ Important:Theforceparameter shouldONLYbe set totruewhen explicitly needed (e.g., "force reindex", "rebuild everything from scratch"). Full reindex is expensive:

- Re-downloads all files from S3
- Regenerates all embeddings
- Rebuilds the entire vector store

For normal operations, always use incremental sync (default behavior).

{ "s3_key": "docs/authentification_magique_symfony.md" }

Retrieves the complete content of a Markdown file from S3 along with metadata:

- Full S3 key: The document's S3 identifier
- Complete Markdown content: Entire document (not chunked)
- Metadata: Size in bytes, last modification date, ETag, chunk count (if indexed)

- View the complete document after finding it viasearch_documentation
- Export documentation for external use
- Understand the full context around a search result
- Display complete documents in third-party integrations

- If a document appears in search results butget_full_documentreturns "not found", it means the file was deleted from S3 after being indexed
- Solution: Runrefresh_indexto synchronize the index with the current S3 state
- The tool will provide a helpful error message indicating when a sync is needed

- resources/list: Lists all indexed Markdown files with metadata (name, URI, size, chunks, last modified)
- resources/read: Reads the full content of a specific file by its URI (e.g.,s3doc://docs/authentication.md)

Use case:When users ask "What files do you have?" or "Show me file X", the LLM can browse and access files directly without semantic search.

This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.

Integrates LLM applications with documentation sources using the Model Context Protocol.

Manage user feedback, roadmap, changelog and user surveys all from your favorite LLM

Share your team's Coding Best Practices with Cursor, VS Code, Claude code, Windsurf, JetBrains IDEs and other coding tools supporting remote MCP connection.

Lance is an agent that runs on its own cloud Mac and handles everything on the App Store Connect side of shipping an iOS app. It connects to coding agents like Claude Code and Codex through MCP and takes over the parts they can't do: creating apps on App Store Connect, setting up notifications / widgets, filling out listing metadata, uploading builds to TestFlight, submitting for review, and responding to App Review feedback. Because it has its own Mac, it can also build, run, and test users' apps and capture screenshots.

An intelligent server for code analysis, collection, and documentation generation using the OpenAI API.

AI-powered semantic search over React documentation for Claude, Cursor, and other MCP clients.

AI-native static site generator with built-in MCP server. Build sites, create content, apply themes, search docs, and deploy via Claude Code or any MCP client.

A universal MCP server that connects to any Storybook site and extracts documentation in real-time using Playwright. Use it with any AI or client that supports MCP (Model Context Protocol)—Cursor, Claude Desktop, Windsurf, or other MCP hosts.

Exposes WebPinch tasks, projects, site audits, and stats as MCP tools/resources/prompts for Claude Code, Cursor, and other MCP clients.