gemini-webapi

gemini-webapi-mcp is a Python-based MCP server that integrates with Google Gemini to generate and edit images, perform file analysis, and chat via Gemini. It authenticates using cookies read from Chrome (via browser-cookie3) or manual cookie values provided through environment variables. The server exposes a set of tools you can invoke through MCP clients: gemini_generate_image for creating or editing images based on prompts, gemini_upload_file for analyzing images, videos, PDFs and documents, gemini_analyze_url for evaluating URLs, gemini_chat and gemini_start_chat for Gemini-powered text conversations, and gemini_reset to reinitialize the client after auth issues. When configured, Claude-like MCP clients can automatically trigger these tools from chat prompts. The server also supports auto-upscaling by requesting a 2x version from Gemini and removes watermark marks using a Reverse Alpha Blending method, all without requiring extra models. To use it, ensure your MCP client is configured to launch the server with the uv command and the proper with-args, then issue prompts like “Generate an image of a cat in watercolor through Gemini” or “Edit this image with a purple hue.”

Prerequisites:

• Python 3.10+ installed on your system
• uv (uvx) runtime available and in PATH
• Git installed

1. Install uv (if not already installed)

# Example using pipx (if you have it)
pipx install uvx
# Or follow uvx install guide for your platform

2. Clone the repository (optional if you already have it)

git clone https://github.com/AndyShaman/gemini-webapi-mcp.git
cd gemini-webapi-mcp

3. Install dependencies and sync (via uv)

uv sync

4. Run the MCP server locally (example)

uv run --with "gemini-webapi-mcp @ git+https://github.com/AndyShaman/gemini-webapi-mcp.git" gemini-webapi-mcp

5. Add MCP config (example for local use)

{
  "mcpServers": {
    "gemini": {
      "command": "uv",
      "args": ["run", "--with", "gemini-webapi-mcp @ git+https://github.com/AndyShaman/gemini-webapi-mcp.git", "gemini-webapi-mcp"]
    }
  }
}

6. Start using Claude Code/Desktop or another MCP client to invoke the Gemini tools through prompts.

Tips and notes:

• If automatic cookie discovery fails, supply cookies manually via environment variables GEMINI_PSID and GEMINI_PSIDTS.
• You can override the Gemini response language with GEMINI_LANGUAGE (default is en).
• For multi-account setups in Chrome, set GEMINI_ACCOUNT_INDEX to choose the account order (0 = first).
• The 2x upscaling relies on Google’s server-side upscaling; if it’s unavailable, the native resolution is used instead.
• Ensure your MCP client supports the standard MCP json config format and points to the correct Python-based server binary via the uv command.
• The watermark removal process uses the Reverse Alpha Blending algorithm; no additional assets or models are needed.
• If you run behind proxies or restricted networks, ensure outbound access to Gemini services is allowed and that cookies can be read from browser storage.