Gemini AI 访问 MCP 服务器

CamoGemini

Gemini AI MCP Server — Access Google Gemini through AI agents via CamoFox anti-detection browser.

Zero API keys. Zero third-party dependencies. Uses browser-based protocol execution (IBPE) — your requests go through a real browser session, not API endpoints.

Quick Start

Prerequisites

CamoFox browser must be running:

docker run -d -p 9377:9377 --name camofox ghcr.io/redf0x1/camofox-browser:latest

Don't have Docker? See CamoFox installation guide.

AI Auto-Setup (Recommended)

Paste this prompt into your AI assistant (Copilot, Claude, Cursor, etc.) and it will set up everything automatically:

Set up CamoGemini MCP server for me. Follow these steps:

1. Check if Docker is available. If yes, check if a container named "camofox" is running:
  - If not running, start it: docker run -d -p 9377:9377 --name camofox ghcr.io/redf0x1/camofox-browser:latest
  - If already running, skip this step

2. Test CamoFox is reachable:
  - curl http://localhost:9377/health
  - If not reachable, wait 5 seconds and retry (CamoFox takes a moment to start)

3. Add the CamoGemini MCP server to my editor config:
  - For VS Code: create/update .vscode/mcp.json with:
    {"servers":{"camo-gemini":{"type":"stdio","command":"npx","args":["-y","camo-gemini@latest"],"env":{"CAMOFOX_URL":"http://localhost:9377"}}}}
  - For Claude Desktop: update ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)
  - For Cursor: update ~/.cursor/mcp.json
  - For Windsurf: update ~/.codeium/windsurf/mcp_config.json
  Use "mcpServers" key instead of "servers" for non-VS Code editors (and omit "type":"stdio").

4. After config is saved, tell me to restart my editor to load the new MCP server.

5. Once restarted, verify the setup by calling these MCP tools in order:
  - gemini_health — should show CamoFox is connected
  - gemini_login — authenticate with Google (account 0)
  - gemini_generate with prompt "Say hello" — should return a response

Report the result of each step.

Install

Option A: npx (Recommended)

No installation needed — add directly to your MCP client config:

{
  "servers": {
    "camo-gemini": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "camo-gemini@latest"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

{
  "mcpServers": {
    "camo-gemini": {
      "command": "npx",
      "args": ["-y", "camo-gemini@latest"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

{
  "mcpServers": {
    "camo-gemini": {
      "command": "npx",
      "args": ["-y", "camo-gemini@latest"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

{
  "mcpServers": {
    "camo-gemini": {
      "command": "npx",
      "args": ["-y", "camo-gemini@latest"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

Option B: From Source

git clone https://github.com/redf0x1/camo-gemini.git
cd camo-gemini
npm install
npm run build

Then use the full path in your MCP config:

{
  "servers": {
    "camo-gemini": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/camo-gemini/dist/index.js"],
      "env": {
        "CAMOFOX_URL": "http://localhost:9377"
      }
    }
  }
}

Setup Wizard

Run the interactive setup wizard to verify your installation:

npx camo-gemini-setup

Verify Installation

Paste this into your AI agent to verify everything works:

Verify my CamoGemini setup:
1) Call gemini_health — is CamoFox connected?
2) Call gemini_login to authenticate with Google
3) Call gemini_generate with prompt "Say hello in 5 words"
Report pass/fail for each step.

MCP Tools (15)

1. gemini_login — Connect to existing CamoFox browser session (session-based, not credential-based)
2. gemini_logout — Disconnect and clean up session
3. gemini_auth_status — Check authentication status
4. gemini_health — Health check across all accounts
5. gemini_generate — Text generation (with auto-delete)
6. gemini_generate_image — Image generation (with multi-account failover + auto-delete)
7. gemini_chat — Multi-turn chat conversation
8. gemini_stream — Legacy buffered text-generation alias (with auto-delete; not incremental streaming)
9. gemini_list_gems — List all Gems (custom + system)
10. gemini_create_gem — Create a custom Gem
11. gemini_update_gem — Update an existing Gem
12. gemini_delete_gem — Delete a Gem
13. gemini_upload_file — Upload a file (base64)
14. gemini_list_accounts — List configured accounts
15. gemini_add_account — Add and authenticate a new account

Environment Variables

Complete runtime environment variables from src/config.ts:

Multi-Account Setup & Failover

• Supports up to 10 accounts (accountIndex 0-9)
• Automatic failover when active account hits retryable limits/errors
• Image generation daily-limit conditions fail over to next healthy account
• Session isolation per account via distinct CamoFox user/browser context

Architecture (IBPE)

MCP Client -> CamoGemini -> CamoFox Browser -> Gemini Web APIs

• All API calls are executed through CamoFox browser context
• Primary request pattern: page.evaluate(() => fetch(url, { credentials: 'include' }))
• Anti-detection is handled at browser engine level (C++ fingerprint spoofing)
• No direct Gemini HTTP client libraries in CamoGemini

Auto-Delete Chat Behavior

Calls to gemini_chat with the same sessionId are processed in order; different sessions can run independently. Clearing a local session/account cancels its queued calls and prevents late results from restoring cleared local state. An explicit account change starts fresh conversation metadata rather than carrying the previous account's conversation into the new account. Multi-turn chat stays pinned to its selected account; choose a different account explicitly to start fresh metadata. One-shot text/image generation retains automatic account failover. Already-invoked generation may still finish; this is not provider-side cancellation or deletion of Gemini history.

• One-shot tools (gemini_generate, gemini_stream, gemini_generate_image) auto-delete conversations by default
• Per-call override: pass autoDelete: false
• Global override: set CAMOGEMINI_AUTO_DELETE_CHAT=false
• Multi-turn gemini_chat does not auto-delete by design

Testing

npm test          # Run all tests
npm run build     # TypeScript build

Docker

Dockerfile is included.

docker build -t camo-gemini .
docker run --rm -it -e CAMOFOX_HOST=host.docker.internal -e CAMOFOX_PORT=9377 camo-gemini

Optional dashboard:

docker run --rm -it -p 9378:9378 -e CAMOGEMINI_DASHBOARD=true camo-gemini

License

MIT