Quick Installation

MCP Server for Claude Code

# Using npm (Recommended)
claude mcp add gemini -s user -- env GEMINI_API_KEY=YOUR_KEY npx -y @rlabs-inc/gemini-mcp

# Using bun
claude mcp add gemini -s user -- env GEMINI_API_KEY=YOUR_KEY bunx @rlabs-inc/gemini-mcp

CLI (Global Install)

# Install globally
npm install -g @rlabs-inc/gemini-mcp

# Set your API key once (stored securely)
gcli config set api-key YOUR_KEY

# Now use any command!
gcli search "latest news"
glci image "sunset over mountains" --ratio 16:9

Get your API key: Visit Google AI Studio - it's free and takes seconds!

Installation Options

# With verbose logging
claude mcp add gemini -s user -- env GEMINI_API_KEY=YOUR_KEY VERBOSE=true bunx -y @rlabs-inc/gemini-mcp

# With custom output directory for generated images/videos
claude mcp add gemini -s user -- env GEMINI_API_KEY=YOUR_KEY GEMINI_OUTPUT_DIR=/path/to/output bunx -y @rlabs-inc/gemini-mcp

Available Tools

gemini-query

Direct queries to Gemini with thinking level control:

prompt: "Explain quantum entanglement"
model: "pro" or "flash"
thinkingLevel: "low" | "medium" | "high" (optional)

• low: Fast responses, minimal reasoning
• medium: Balanced (Flash only)
• high: Deep reasoning for complex tasks (default)

gemini-generate-image

Generate images with Nano Banana Pro (Claude can SEE them!):

prompt: "a futuristic city at sunset"
style: "cyberpunk" (optional)
aspectRatio: "16:9" (1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9)
imageSize: "2K" (1K, 2K, 4K)
useGoogleSearch: false (ground in real-world info)
thinkingLevel: "high" (optional - minimal, low, medium, high)
personGeneration: "ALLOW_ALL" (optional - ALLOW_ALL, ALLOW_ADULT, ALLOW_NONE)
seed: 42 (optional - for reproducible results)

gemini-start-image-edit

Start a multi-turn image editing session:

prompt: "a cozy cabin in the mountains"
aspectRatio: "16:9"
imageSize: "2K"
useGoogleSearch: false
thinkingLevel: "high" (optional - minimal, low, medium, high)
personGeneration: "ALLOW_ALL" (optional - ALLOW_ALL, ALLOW_ADULT, ALLOW_NONE)
seed: 42 (optional - for reproducible results)

Returns a session ID for iterative editing.

gemini-continue-image-edit

Continue refining an image:

sessionId: "edit-123456789"
prompt: "add snow on the roof and make it nighttime"

gemini-end-image-edit

Close an editing session:

sessionId: "edit-123456789"

gemini-list-image-sessions

List all active editing sessions.

gemini-generate-video

Generate videos using Veo:

prompt: "a cat playing piano"
aspectRatio: "16:9" (optional)
negativePrompt: "blurry, text" (optional)

Video generation is async (takes 1-5 minutes). Use gemini-check-video to poll.

gemini-check-video

Check video generation status and download when complete:

operationId: "operations/xxx-xxx-xxx"

gemini-analyze-code

Analyze code for issues:

code: "function foo() { ... }"
language: "typescript" (optional)
focus: "quality" | "security" | "performance" | "bugs" | "general"

gemini-analyze-text

Analyze text content:

text: "Your text here..."
type: "sentiment" | "summary" | "entities" | "key-points" | "general"

gemini-brainstorm

Collaborative brainstorming:

prompt: "How could we implement real-time collaboration?"
claudeThoughts: "I think we should use WebSockets..."
maxRounds: 3 (optional)

gemini-summarize

Summarize content:

content: "Long text to summarize..."
length: "brief" | "moderate" | "detailed"
format: "paragraph" | "bullet-points" | "outline"

gemini-run-code

Let Gemini write and execute Python code:

prompt: "Calculate the first 50 prime numbers and plot them"
data: "optional CSV data to analyze" (optional)

Supports libraries: numpy, pandas, matplotlib, scipy, scikit-learn, tensorflow, and more. Generated charts are saved to the output directory and returned as images.

gemini-search

Real-time web search with citations:

query: "What happened in tech news this week?"
returnCitations: true (default)

Returns grounded responses with inline citations and source URLs.

gemini-structured

Get JSON responses matching a schema:

prompt: "Extract the meeting details from this email..."
schema: '{"type":"object","properties":{"date":{"type":"string"},"attendees":{"type":"array"}}}'
useGoogleSearch: false (optional)

gemini-extract

Convenience tool for common extraction patterns:

text: "Your text to analyze..."
extractType: "entities" | "facts" | "summary" | "keywords" | "sentiment" | "custom"
customFields: "name, date, amount" (for custom extraction)

gemini-youtube

Analyze YouTube videos directly:

url: "https://www.youtube.com/watch?v=..."
question: "What happens at 2:30?"
startTime: "1m30s" (optional, for clipping)
endTime: "5m00s" (optional, for clipping)

gemini-youtube-summary

Quick video summarization:

url: "https://www.youtube.com/watch?v=..."
style: "brief" | "detailed" | "bullet-points" | "chapters"

gemini-analyze-document

Analyze PDFs and documents:

filePath: "/path/to/document.pdf"
question: "Summarize the key findings"
mediaResolution: "low" | "medium" | "high"

gemini-summarize-pdf

Quick PDF summarization:

filePath: "/path/to/document.pdf"
style: "brief" | "detailed" | "outline" | "key-points"

gemini-extract-tables

Extract tables from documents:

filePath: "/path/to/document.pdf"
outputFormat: "markdown" | "csv" | "json"

Workflow: Claude + Gemini

The killer combination for development:

| Claude | Gemini | |--------|--------| | Complex logic | Frontend/UI | | Architecture | Visual components | | Backend code | Image generation | | Integration | React/CSS styling | | Reasoning | Creative generation |

Example workflow:

1. Ask Claude to design the backend API
2. Use gemini-generate-image for UI mockups
3. Ask Gemini to generate React components via gemini-query
4. Use multi-turn editing to refine visuals
5. Let Claude wire everything together

Environment Variables

| Variable | Required | Default | Description | |-------------------------|----------|------------------------------|-------------------------------| | GEMINI_API_KEY | Yes | - | Your Google Gemini API key | | GEMINI_OUTPUT_DIR | No | ./gemini-output | Where to save generated files | | GEMINI_MODEL | No | - | Override model for init test | | GEMINI_PRO_MODEL | No | gemini-3-pro-preview | Pro model (Gemini 3) | | GEMINI_FLASH_MODEL | No | gemini-3-flash-preview | Flash model (Gemini 3) | | GEMINI_IMAGE_MODEL | No | gemini-3-pro-image-preview | Image model (Nano Banana Pro) | | GEMINI_IMAGE_THINKING_LEVEL | No | high | Default thinking level for image generation (minimal, low, medium, high) | | GEMINI_VIDEO_MODEL | No | veo-2.0-generate-001 | Video model | | VERBOSE | No | false | Enabl