gemini

gemini-mcp

Claude Code plugin that bridges to Google Gemini CLI for multimodal tasks.

Claude Code is powerful at coding and reasoning, but weaker at image/vision tasks. This MCP server lets Claude automatically delegate multimodal work to Gemini — the best of both worlds.

Tools

Tool	Description
gemini_prompt	Send a text prompt to Gemini
gemini_vision	Analyze images using Gemini's multimodal capabilities
gemini_code	Delegate code review or analysis to Gemini
gemini_summarize	Summarize long text or files

All tools support optional parameters:

• timeout — Timeout in seconds (default: 120)
• language — Response language: "ko", "en", "ja", "zh", or any language name. Use "none" to skip

Note: Uses gemini-3-flash-preview model (the only model currently available in Gemini CLI).

Supported Platforms

OS	Status	Notes
macOS	Fully supported	Primary development platform
Linux	Fully supported
Windows	Fully supported	Requires Bun and Gemini CLI on PATH

Prerequisites

• Bun runtime (macOS, Linux, Windows)
• Gemini CLI installed and authenticated

# Install Gemini CLI
npm install -g @google/gemini-cli

# Authenticate (run once)
gemini

Installation

Quick Install (Recommended)

Run the installer script — it checks prerequisites, clones the repo, and configures Claude Desktop:

bash <(curl -fsSL https://raw.githubusercontent.com/netwaif/gemini-mcp/main/install.sh)

Or if you already cloned the repo:

bash install.sh

Tip: You can also ask Claude to install it for you: "https://github.com/netwaif/gemini-mcp 이 플러그인을 설치해줘"

Manual Install — Claude Code CLI

Clone this repo into your Claude Code plugins directory:

macOS / Linux:

mkdir -p ~/.claude/plugins/local
cd ~/.claude/plugins/local
git clone https://github.com/netwaif/gemini-mcp.git

Windows (PowerShell):

mkdir -Force "$env:USERPROFILE\.claude\plugins\local"
cd "$env:USERPROFILE\.claude\plugins\local"
git clone https://github.com/netwaif/gemini-mcp.git

Restart Claude Code — the plugin will be auto-discovered.

As a standalone MCP server

Add to your MCP config (~/.mcp.json on macOS/Linux, %USERPROFILE%\.mcp.json on Windows):

{
  "mcpServers": {
    "gemini": {
      "command": "bun",
      "args": ["run", "--cwd", "/path/to/gemini-mcp", "--shell=bun", "--silent", "start"]
    }
  }
}

With Claude Desktop App

Claude Desktop uses its own config file. Edit the following file (create it if it doesn't exist):

macOS:

~/Library/Application Support/Claude/claude_desktop_config.json

Windows:

%APPDATA%\Claude\claude_desktop_config.json

Add the Gemini MCP server config — note that you must use an absolute path (not ${CLAUDE_PLUGIN_ROOT}):

macOS / Linux:

{
  "mcpServers": {
    "gemini": {
      "command": "bun",
      "args": ["run", "--cwd", "/absolute/path/to/gemini-mcp", "--shell=bun", "--silent", "start"]
    }
  }
}

Windows:

{
  "mcpServers": {
    "gemini": {
      "command": "bun",
      "args": ["run", "--cwd", "C:\\Users\\YourName\\.claude\\plugins\\local\\gemini-mcp", "--shell=bun", "--silent", "start"]
    }
  }
}

Tip: If bun is not found, use the full path (e.g., /opt/homebrew/bin/bun on macOS or C:\Users\YourName\.bun\bin\bun.exe on Windows).

Restart the Claude Desktop app to load the new MCP server.

Skill Installation (Claude Desktop App)

The MCP tools work without the skill, but importing the /gemini skill enables Claude to invoke Gemini more reliably.

1. Update Claude Desktop App to the latest version
2. In the left sidebar: Customize → Skills → + → Upload Skill
3. Select the gemini.skill file
   • If you ran the installer, it's copied to ~/Desktop/gemini.skill
   • Otherwise, find it at ~/.claude/plugins/local/gemini-mcp/gemini.skill
4. Restart the Claude Desktop App

The gemini.skill file is included in this repository. It is a zip archive containing SKILL.md. After importing, you can delete the copy from your Desktop.

How it works

Claude Code  →  MCP Server (this)  →  Gemini CLI  →  Gemini API
                (stdio transport)      (subprocess)

1. Claude decides it needs Gemini (e.g., for image analysis)
2. Calls the appropriate MCP tool (gemini_vision, etc.)
3. This server spawns gemini --approval-mode yolo -p "..." --model=...
4. Returns Gemini's response back to Claude

Environment Variables

Variable	Default	Description
GEMINI_PATH	gemini	Path to the Gemini CLI binary

License

MIT