Skip to content

Gemini Bridge Plugin

Model-agnostic bridge to Google Gemini 2.5 Pro for Claude Code. Enables long-context codebase analysis (1M tokens), multimodal vision, and independent model validation within your existing Claude Code workflow.

Version 1.0.1

Gemini runs as an MCP tool — Claude calls it exactly like any other tool. No second terminal, no context switch, no workflow disruption.

Commands

/gemini-bridge:analyze

Delegate text or code analysis to Gemini's long-context model.

Features:

  • 📝 Text prompts and second opinions
  • 📦 Large codebase analysis (up to 1M tokens)
  • 🔍 Security audits across entire repositories
  • 🧠 Independent model validation

Usage:

Bash
1
2
/gemini-bridge:analyze Perform a full security audit of the src/ directory
/gemini-bridge:analyze Review this architecture for scalability issues

/gemini-bridge:compare

Compare two technical approaches with Gemini as a neutral reviewer.

Features:

  • ⚖️ Neutral comparison of two options
  • 📊 Structured evaluation with criteria
  • 🎯 Independent assessment (no anchoring to Claude's preference)

Usage:

Bash
1
2
/gemini-bridge:compare Should we use event sourcing or CQRS for the exam engine?
/gemini-bridge:compare Redis vs PostgreSQL for session storage

/gemini-bridge:vision

Convert screenshots, diagrams, or PDFs into code or documentation.

Features:

  • 🖼️ Screenshot to component conversion
  • 📐 Diagram analysis and documentation
  • 📄 PDF content extraction
  • 🎨 UI mockup to React/Tailwind code

Usage:

Bash
1
2
/gemini-bridge:vision ./designs/dashboard-mockup.png "Generate a React + Tailwind component"
/gemini-bridge:vision ./docs/architecture.pdf "Summarize the key design decisions"

Agents

All agents activate automatically based on context. See Skills & Agents Activation Guide for details.

Gemini Analyst

Sub-agent for delegating analysis tasks to Gemini.

Activation: - Automatic: When tasks require long-context analysis or multimodal processing - Manual: "Use gemini-analyst to review this codebase"

Expertise: - 📦 Large codebase analysis - 🖼️ Multimodal vision tasks - 🔍 Security and architecture reviews - 📊 Structured report generation

Model Router

Orchestrator agent with capability-based routing logic.

Activation: - Automatic: When tasks benefit from multi-model collaboration - Manual: "Route this task to the best model"

Routing Logic:

Text Only
1
2
3
4
5
6
7
Task                        → Model       → Reason
────────────────────────────────────────────────────
Entire repo analysis        → Gemini      1M context window
Write auth middleware       → Claude      Tool use expertise
Compare two approaches      → Gemini      Neutral review
Generate tests              → Claude      Code quality focus
Convert mockup to component → Gemini      Vision capability

Skills

Gemini Analyst Skill

Background knowledge for Claude on when and how to delegate to Gemini effectively.

MCP Tools

The following tools are available to all Claude agents when the plugin is installed:

Tool Purpose
gemini_status Check bridge connectivity and model info
gemini_analyze_text Text prompts and second opinions
gemini_analyze_codebase Large codebase analysis (up to 1M tokens)
gemini_analyze_image Screenshot, diagram, and PDF analysis
gemini_compare_approaches Neutral comparison of two technical options

Installation

1. Enable the Plugin

JSON
1
2
3
4
5
{
  "enabledPlugins": {
    "gemini-bridge@talent-factory": true
  }
}

2. Install Python Dependencies

Bash
1
2
3
pip install google-genai fastmcp
# or with uv (recommended):
uv add google-genai fastmcp

3. Set API Key

Bash
1
2
# Get your key at: https://aistudio.google.com/app/apikey
export GEMINI_API_KEY="your-key-here"

4. Verify Connection

Bash
1
/gemini-bridge:analyze Check if the bridge is working

Configuration

Custom Model Version

Bash
1
2
export GEMINI_MODEL="gemini-2.5-flash-preview-04-17"  # Flash for cost savings
export GEMINI_MODEL="gemini-2.5-pro"                   # Default (Pro)

Temperature Tuning

Python
1
2
gemini_analyze_text(prompt="...", temperature=0.7)  # More creative
gemini_analyze_text(prompt="...", temperature=0.0)  # Deterministic

Security and Privacy

  • API key: Stored as environment variable, never committed to git
  • Data: Code content is sent to Google's Gemini API — review your data policies
  • .gitignore: Ensure .env files with your API key are excluded

Use Cases

Large Codebase Security Review

Bash
1
/gemini-bridge:analyze Perform a full security audit of the src/ directory

Claude collects all source files, delegates to Gemini's 1M-token context, and returns a structured security report.

Architecture Decision with Independent Review

Bash
1
/gemini-bridge:compare Should we use event sourcing or CQRS for the exam engine?

Claude prepares both approaches; Gemini evaluates them neutrally.

UI Screenshot to Component

Bash
1
/gemini-bridge:vision ./designs/dashboard-mockup.png "Generate a React + Tailwind component"

Teaching: Model-Agnostic AI Engineering

This plugin demonstrates the model-agnostic pattern in AI-Assisted Software Engineering courses:

  1. Separation of Concerns — Routing logic in config, not code
  2. Capability-Based Routing — Each model handles what it does best
  3. Multi-Model Validation — Claude proposes, Gemini validates, Claude synthesizes

Changelog

Version 1.0.0 (2026-03-13)

  • Initial release with 3 commands, 2 agents, 1 skill, and 5 MCP tools
  • Google Gemini 2.5 Pro integration via google-genai SDK
  • Multimodal vision support (images, PDFs)
  • Configurable model via GEMINI_MODEL environment variable
  • Comprehensive test suite (29 tests)

License

MIT License - see LICENSE file for details.

Made with care by Talent Factory GmbH