GeminiMCP

An MCP server that exposes Google Gemini models to MCP clients (Claude Code, IDE extensions, custom tooling).

What it does

• gemini_ask — multimodal question answering with composable GitHub context (PRs, commits, diffs, files)
• gemini_search — Google Search-grounded queries returning structured JSON
• 3 workflow prompts — review_pr, explain_commit, compare_refs
• 8 coding prompts — code review, explain, debug, refactor, architecture, tests, security, research
• Gemini 3-style XML envelope — every user turn is rendered as <context> / <task> / <final_instruction> with attacker-controllable bodies emitted as raw text
• Server-side system-prompt selection — a Flash classifier picks a category-specific SystemInstruction and matching <final_instruction> body
• Two transports: HTTP (JWT-secured, preferred) and stdio (local fallback)

Quick start

# Build
go build -o bin/mcp-gemini .

# Configure
cp .env.example .env
# Edit .env — set GEMINI_API_KEY at minimum

# Run (stdio, for Claude Code)
./bin/mcp-gemini

# Run (HTTP)
./bin/mcp-gemini --transport=http

Claude Code mcpServers entry (stdio):

{
  "gemini": {
    "command": "/path/to/bin/mcp-gemini",
    "env": { "GEMINI_API_KEY": "your-key" }
  }
}

Documentation

Doc	Contents
docs/QUICKREF.md	One-page cheat sheet — parameters, env vars, CLI flags, examples
docs/TOOLS.md	Full tool and prompt reference with all parameters and examples
docs/PROMPTS.md	MCP-prompt registry and server-side system-prompt selection
docs/USAGE.md	Installation, transport modes, authentication, all env vars
docs/ARCHITECTURE OVERVIEW.md	Codebase map, component diagram, request lifecycle
.env.example	Annotated configuration template

Development

./run_test.sh     # tests
./run_lint.sh     # golangci-lint
./run_format.sh   # goimports + gofmt

Requirements

• Go 1.25+
• GEMINI_API_KEY (Google AI Studio)
• GEMINI_GITHUB_TOKEN (GitHub PAT — only for github_* parameters)