antigravity-mcp-bridge
Enables programmatic control of Google Antigravity 2.0 agents via MCP, with secure HTTP/stdio transports and Tailscale Funnel for remote access.
README
Antigravity MCP Bridge (Antigravity 2.0 ↔ Gemini Spark)
Programmatically control Google Antigravity 2.0 agents on Windows via Gemini Spark and remote/local MCP clients over Streamable HTTP and stdio transport.
Architecture Overview
Gemini Spark (or Remote MCP Client)
│
│ HTTPS Request (Authorization: Bearer <MCP_AUTH_TOKEN>)
▼
Tailscale Funnel (Public HTTPS: https://<node>.<tailnet>.ts.net/mcp)
│
│ Encrypted Local Tunnel
▼
Streamable HTTP MCP Server (src/http_server.py on 127.0.0.1:8000)
│
├── AuthMiddleware (Validates MCP_AUTH_TOKEN -> 401 Unauthorized if invalid)
├── /health & /healthz Diagnostic Endpoints
│
▼
FastMCP Core Server (src/server.py - Streamable HTTP & stdio)
│
├── launch_antigravity_agent
├── get_session_status
├── list_workspace_artifacts
└── run_antigravity_cli_command
│
▼
Antigravity Bridge Adapter -> agentapi.bat -> Antigravity 2.0 Runtime -> Local Workspace
Quick Start (Windows One-Click Operations)
START.bat: Initializes environment, creates.envwith secure auth token, starts background HTTP server on127.0.0.1:8000, enables Tailscale Funnel, and displays connection banner.STATUS.bat: Inspects live server health, PID status, auth token state, Tailscale login & Funnel status, and Antigravity runtime availability.STOP.bat: Gracefully terminates the HTTP MCP server process and disables Tailscale Funnel mapping.
Key Features
- Dual Transport Support:
- Stdio transport for local MCP clients (
python src/server.py). - Streamable HTTP transport for Gemini Spark and remote clients (
START.bat/python -m src.http_server).
- Stdio transport for local MCP clients (
- Authentication & Token Protection:
AuthMiddlewareenforcing token validation on/mcp.- Supports
Authorization: Bearer <token>,x-api-key: <token>, or?token=<token>. - Auto-generates 32-character hex token if missing in
.env.
- Tailscale Funnel Integration:
- Secure public HTTPS endpoint without purchasing a domain.
- Robust detection, status reporting, and resolution guidance when Tailscale is absent or disabled.
- Antigravity 2.0 Local Engine:
- Zero-config integration with local Antigravity Language Server via
agentapi.bat.
- Zero-config integration with local Antigravity Language Server via
- Safety & Workspace Isolation:
- Workspace boundary validation preventing path traversal attacks.
- Automatic secret key and token redaction from output streams.
- Concurrent workspace execution locking preventing multi-agent conflicts.
Exposed MCP Tools
launch_antigravity_agent: Launches an Antigravity agent task targeting a local workspace.get_session_status: Inspects session status, timestamps, and live metadata.list_workspace_artifacts: Indexes files, sizes, timestamps, and SHA256 hashes in workspace.run_antigravity_cli_command: Executes supportedagentapisubcommands (get-conversation-metadata,send-message,new-conversation).
Verification & Test Suite
Run full automated unit and integration tests:
.venv\Scripts\python.exe -m pytest tests/
Run real end-to-end HTTP MCP integration test:
.venv\Scripts\python.exe remote_smoke_test.py
Run local stdio smoke test:
.venv\Scripts\python.exe smoke_test.py
Documentation
Detailed documentation is available in docs/:
docs/remote-access.md: Complete guide for remote access, Tailscale Funnel, and Gemini Spark connection.docs/mcp-setup.md: Client configuration snippets for Gemini Spark and Antigravity IDE.docs/security.md: Security model, path validation, and secret redaction rules.docs/troubleshooting.md: Troubleshooting matrix and diagnostic recovery steps.FINAL-REPORT.md: Complete integration status report and empirical verification matrix.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。