Bookmark Geni MCP Server

Stand-alone MCP (Model Context Protocol) server for processing local browser bookmarks and making url/bookmarks easily searchable using natural language.

<video controls src="./recording1.mov" title="Bookmark Geni MCP Demo"></video>

Connect to Claud Desk top or Gemini CLI <img width="832" height="595" alt="Screenshot 2025-11-28 at 10 16 08PM" src="https://github.com/user-attachments/assets/99c368bd-127e-4f08-8e04-0fbfd7322146" />

<img width="835" height="614" alt="Screenshot 2025-11-28 at 10 19 21PM" src="https://github.com/user-attachments/assets/978e519a-4327-44d7-a8a6-31d250ebcec9" /> <img width="842" height="606" alt="Screenshot 2025-11-28 at 10 19 49PM" src="https://github.com/user-attachments/assets/8be42616-c138-4d98-81ee-0765c0ddfc73" />

Features

• Multi-Browser Support: Reads bookmarks from Chrome, Edge, Firefox, Opera, ChatGPT Atlas, and Perplexity Comet
• Content Extraction: Fetches HTML content from URLs and extracts text for semantic search
• Metadata Extraction: Extracts descriptions from HTML metadata tags (Open Graph, meta description, title)
• Vector Storage: Stores bookmark embeddings in ChromaDB using sentence transformer model (all-MiniLM-L6-v2)
• RAG Search: Query bookmarks using natural language with metadata filtering
• Portability: Export and import embeddings to/from pickle files for easy transfer
• Performance: Batch processing with concurrency and caching

Installation

1. Install dependencies:

   pip install -r requirements.txt
   

2. Configure the server by editing config.yaml (optional - defaults are provided)

3. Make start script executable:

   chmod +x scripts/start_mcp_server.sh
   

Usage

Once the server is started, it can be used with any MCP client. To index all browser bookmarks and generate metadata, run the following command:

"Generate metadata for chrome bookmarks"

Once all bookmarks are indexed, you can query them using the following command:

"Query bookmarks for 'python'"

Standalone MCP Server

The server can be used independently with any MCP client by referencing mcp.json:

# Start the server using the bash script
./scripts/start_mcp_server.sh

With Gemini CLI

Connect the server to the Gemini CLI using the following command:

# Add the server configuration to your Gemini CLI settings
# Edit ~/.gemini/settings.json and add the following to the "mcpServers" section:

{
  "mcpServers": {
    "bookmarkGeni": {
      "command": "bash",
      "args": ["/path/to/bookmark_geni_mcp/scripts/start_mcp_server.sh"],
      "env": {
        "PYTHON_PATH": "/path/to/your/python3"
      }
    }
  }
}

# Or use the provided mcp.json as a reference for the configuration

With Calude Desktop

Connect the server to the Calude Desktop by adding the following to the Calude Desktop settings:

{
  "mcpServers": {
    "bookmarkGeni": {
      "command": "bash",
      "args": ["/path/to/bookmark_geni_mcp/scripts/start_mcp_server.sh"],
      "env": {
        "PYTHON_PATH": "/usr/bin/python3"
      }
    }
  }
}

Note: Replace /path/to/bookmark_geni_mcp with the actual path to this repository and /path/to/your/python3 with your Python interpreter path.

Configuration

The server reads configuration from config.yaml in the MCP server root directory. This includes:

• Browser enable/disable settings
• ChromaDB path (relative to MCP server root or absolute path)
• Metadata JSONL path (relative to MCP server root or absolute path)
• URL processing limit (default: -1, meaning process all URLs)
• Debug mode

Example config.yaml:

debug: false
browsers:
  Chrome:
    enabled: true
  Edge:
    enabled: true
    # Optional: Override default path detection
    # paths:
    #   - "/path/to/custom/Bookmarks"
chromaDbPath: ".chromadb"
metadataJsonlPath: "data/bookmarks_metadata.jsonl"
urlLimit: -1  # -1 means process all, set to positive number to limit

The server is now completely independent and does not require the Gemini CLI extension folder.

Browser Support

The server supports the following browsers:

• Chrome: Windows, macOS, Linux
• Edge: Windows, macOS, Linux
• Firefox: Windows, macOS, Linux
• Opera: Windows, macOS, Linux
• ChatGPT Atlas: macOS (Chromium-based)
• Perplexity Comet: Windows, macOS, Linux (Chromium-based)

Note: Safari is not supported because reading Bookmarks.plist requires special macOS permissions that are not granted by default. To use Safari bookmarks, you would need to grant Full Disk Access permissions to the Python interpreter, which is not recommended for security reasons.

Tools

The server provides the following MCP tools:

1. generate_bookmarks_metadata

   • Scans selected browsers for bookmarks
   • Fetches HTML content and generates metadata
   • Creates embeddings and stores them in ChromaDB
   • Parameters: browsers (e.g., "Chrome,Safari" or "All")

2. query_bookmarks

   • Performs semantic search on stored bookmarks
   • Supports metadata filtering
   • Parameters:
     • query: Search text
     • limit: Max results (default 10)
     • where: Filter dict (e.g., {"folder": "Work"})

3. list_browsers

   • Lists installed browsers and their detected bookmark file paths
   • Parameters: None

4. get_stats

   • Returns database statistics (total count, collection info)
   • Parameters: None

5. export_embeddings

   • Exports all data to a pickle file for backup or transfer
   • Parameters: pickle_path (optional)

6. import_embeddings

   • Imports data from a pickle file
   • Parameters: pickle_path (required)

See mcp.json for detailed schema definitions.

Workflow

flowchart TD
    %% Nodes
    User([User / CLI])
    Server[MCP Server]
    Detector[Browser Detector]
    Parser[Bookmark Parser]
    Generat…