---

A personal collection of MCP servers I built for my Claude Code workflow. Some wrap heavier upstream servers to reduce token usage, others add capabilities that don't exist elsewhere.

Tools

Hub	Purpose	Backend	Notes
agent-mesh	Headless agent runners	Claude/Codex/Gemini CLIs	Multi-agent orchestration
browser-hub	Browser automation	agent-browser	24x token reduction, batch form filling
search-hub	Web research	OpenAI API	Synthesized answers with citations
image-hub	Image generation	Vertex AI Gemini	Gemini 3 Pro native image gen
gdrive-hub	Google Drive access	Google Drive API	Service account auth, Shared Drives
google-ads-hub	Google Ads analysis	Google Ads API	Read-only reporting tools

Installation

Each tool is a standalone Python package managed with uv:

cd browser-hub && uv sync   # Browser automation
cd search-hub && uv sync    # Web research
cd image-hub && uv sync     # Image generation
cd gdrive-hub && uv sync    # Google Drive access

Configuration

Add to ~/.claude.json:

{
  "mcpServers": {
    "browser-hub": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "--directory", "/path/to/mcp-tools/browser-hub", "browser-hub"]
    },
    "search-hub": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "--directory", "/path/to/mcp-tools/search-hub", "search-hub"],
      "env": {
        "OPENAI_API_KEY": "${OPENAI_API_KEY}"
      }
    },
    "image-hub": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "--directory", "/path/to/mcp-tools/image-hub", "image-hub"],
      "env": {
        "GOOGLE_CLOUD_PROJECT": "${GOOGLE_CLOUD_PROJECT}"
      }
    },
    "gdrive-hub": {
      "type": "stdio",
      "command": "uv",
      "args": ["run", "--directory", "/path/to/mcp-tools/gdrive-hub", "gdrive-hub"],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account.json",
        "GDRIVE_DRIVE_ID": "your-shared-drive-id"
      }
    }
  }
}

Registry (optional, easier for agents)

To avoid editing large config files by hand, use the central registry:

cd ~/git/mcp-tools
python registry/sync_mcp.py        # dry-run
python registry/sync_mcp.py --diff # show changes
python registry/sync_mcp.py --write

Source of truth: registry/mcp-registry.toml. Local overrides: registry/mcp-registry.local.toml (not committed).

Quick Start

Browser Automation

browser(action="navigate", url="https://example.com")
browser(action="look")  # Screenshot + filtered elements in ONE call

# Batch fill forms (10x faster than individual type calls)
browser(action="fill_form", fields={
    "#email": "user@example.com",
    "#name": "John Doe",
    "#country": "United States"
})

browser(action="click", ref="@e5")  # Refs from accessibility tree

Web Research

# Ask complete questions, not keywords
web_research(task="What are the latest developments in quantum computing?")
web_research(task="Compare React vs Vue", reasoning_effort="high")

Image Generation

# Generate images with Gemini
generate_image(prompt="A sunset over mountains", output_path="/tmp/sunset.png")

Google Drive

# List files in root (or specific folder)
list_files()
list_files(folder_id="1ABC...")

# Search with Drive query syntax
list_files(query="name contains 'report'")
list_files(query="mimeType = 'application/pdf'")
list_files(query="modifiedTime > '2024-01-01'")

# Upload and download
upload_file(local_path="/tmp/doc.pdf", folder_id="1ABC...")
download_file(file_id="1XYZ...", export_format="pdf")  # Exports Google Docs

Requirements

Dependency	Purpose
Python 3.11+	Runtime
uv	Package management
Claude/Codex/Gemini CLIs	agent-mesh
agent-browser	browser-hub (npm i -g agent-browser && agent-browser install)
OpenAI API key	search-hub
Google Cloud project	image-hub
Google service account	gdrive-hub (with Drive API access)
Google Ads API credentials	google-ads-hub

License

MIT