---

Overview

Google Gemini cloud services connect to external MCP servers over public HTTPS URLs. This server exposes an HTTP/SSE interface for Gemini while providing direct access to the local Antigravity IDE workspace, including:

• Discovery and inspection of 160+ skills across workspace, global, and plugin scopes.
• Inspection of plugins (cursor-team-kit, pstack, ponytail), rules (GEMINI.md, AGENTS.md), and transcripts.
• Command execution within the active workspace directory.
• Dynamic workspace switching at runtime (set_active_workspace) without server restarts.
• Zero external Python dependencies: implemented with Python's standard library.

---

Quick Start

1. Install from PyPI (Recommended)

Install using pip or execute on-the-fly with uvx / pipx:

# Standard installation
pip install antigravity-mcp

# Or run instantly without pre-installing:
uvx antigravity-mcp --tunnel

2. Install from Source

Clone the repository and execute the setup script:

git clone https://github.com/haydermuhib/gemini-antigravity-orchestrator-mcp.git
cd gemini-antigravity-orchestrator-mcp
chmod +x install.sh && ./install.sh

The script verifies Python 3.10+, checks for cloudflared, installs the antigravity-mcp CLI into ~/.local/bin, and registers the server with local Antigravity IDE configuration.

---

Prerequisites

• Python 3.10 or higher
• Cloudflared (creates the HTTPS tunnel for Gemini Connected Apps)
  • Linux: sudo apt install cloudflared or download from Cloudflare
  • macOS: brew install cloudflared
  • Windows: winget install Cloudflare.cloudflared

---

Running the Server

Background Service (Linux and macOS)

# Start background server and tunnel
antigravity-mcp start   # or ./start_daemon.sh

# View status and retrieve the active public URL
antigravity-mcp status  # or ./status.sh

# Stop background server
antigravity-mcp stop    # or ./stop_daemon.sh

Foreground Mode

Using Pixi:

pixi run tunnel

Using standard Python:

# Terminal 1: run server on port 8080
python3 server.py --http 8080

# Terminal 2: open tunnel
cloudflared tunnel --url http://127.0.0.1:8080

On Windows (Command Prompt or PowerShell):

python server.py --tunnel

---

Connecting to Google Gemini Connected Apps

1. Start the server using one of the methods above.
2. Run ./status.sh or copy the HTTPS tunnel URL:

   https://<subdomain>.trycloudflare.com/sse
   

3. In Google Gemini, navigate to Settings > Connected Apps > Connect to an MCP server.
4. Configure the connection fields:
   • MCP server URL: https://<subdomain>.trycloudflare.com/sse
   • Client ID: gemini-client (or any string)
   • Client secret: gemini-secret (or any string)
5. Click Connect. Gemini validates the endpoint and loads the tool catalog.

---

Local Antigravity IDE Configuration (Stdio)

To configure the server inside Google Antigravity IDE, Claude Desktop, or Cursor, add it to your mcp_config.json:

Option A: Direct PyPI Execution via uvx (No clone required)

{
  "mcpServers": {
    "antigravity-orchestrator": {
      "command": "uvx",
      "args": [
        "antigravity-mcp",
        "--stdio"
      ]
    }
  }
}

Option B: Local Source / Development

{
  "mcpServers": {
    "antigravity-orchestrator": {
      "command": "python3",
      "args": [
        "/path/to/gemini-antigravity-orchestrator-mcp/server.py",
        "--stdio"
      ],
      "env": {
        "PYTHONUNBUFFERED": "1"
      },
      "disabled": false
    }
  }
}

---

Available Tools

The server provides 18 tools to connected models:

Tool Name	Description	Key Arguments
get_antigravity_status	Returns IDE status, workspace path, tool/rule counts, and runtime info.	None
list_known_workspaces	Lists detected workspaces on the machine and shows which is active.	None
set_active_workspace	Switches the active workspace to another project directory at runtime.	workspace_path (required)
list_skills	Discovers skills across workspace, global, builtin, and plugin scopes.	scope, plugin_filter
get_skill	Retrieves SKILL.md text and metadata for a given skill.	skill_name (required)
search_skills	Searches skills using keyword matching.	query (required)
list_plugins	Lists installed plugins, manifests, versions, and subagents.	None
get_plugin_details	Returns full details of a plugin and its configuration.	plugin_name (required)
list_rules	Discovers active rules in workspace and global settings.	None
get_rule	Reads the content and trigger conditions of a rule.	rule_identifier (required)
list_conversations	Lists recent conversation IDs and brain artifacts.	None
get_conversation_transcript	Reads recent steps, user inputs, and tool calls from transcript.jsonl.	conversation_id, limit
list_artifacts	Lists markdown reports and diffs in the conversation brain.	conversation_id
read_artifact	Reads the markdown content of an artifact file.	artifact_name_or_path (required)
execute_workspace_command	Executes shell commands in the workspace with exit code, stdout, and stderr.	command (required), cwd, timeout
create_workspace_skill	Scaffolds a new Antigravity skill directory and SKILL.md.	name, description, instructions, scope
create_workspace_rule	Creates or updates an Antigravity rule in .agents/rules/.	filename, content, scope, trigger
orchestrate_agent_workflow	Generates an execution plan matching task intent with skills and rules.	task (required), preferred_aesthetic

---

Protocol Transports

1. Server-Sent Events (SSE): GET /sse and POST /message?sessionId=<id>
2. Streamable HTTP POST: POST / and POST /mcp
3. Stdio Protocol: python3 server.py --stdio

---

Testing

Run the automated test suites:

# HTTP, SSE, OAuth, HEAD pre-flight, and static asset tests
python3 test_http_mcp.py

# Stdio JSON-RPC 2.0 protocol and 18-tool verification
python3 test_mcp.py

---

License

MIT License. See LICENSE for details.