Flow MCP: Native Google Flow AI Integration

This repository contains the configuration and guides to run and connect the Google Flow MCP server remotely.

---

📂 Files in this Repository

• docker-compose.yml — Configuration to run the Flow API & MCP server container.
• mcp — Configuration file to connect local Claude Desktop/Cursor to the Docker container.

---

🚀 1. Deploying the Server (On Any PC)

To start the server:

1. Ensure Docker is running.
2. Run this command in the repository folder:

   docker-compose up -d

3. Load Chrome and verify that your Flow Chrome Extension is connected.

Note: If you deploy on another PC with a different Cloudflare subdomain, update the PUBLIC_BASE_URL in docker-compose.yml before starting.

---

🔌 2. Connecting to Cursor / Desktop Apps (SSE Transport)

To connect your IDE or desktop app (like Cursor) to the server over the internet:

Open Cursor Settings (Cmd + ,) -> Models -> MCP -> Add New MCP Server:

• Name: flow
• Type: SSE
• URL: https://flow.chatbulky.com/sse

---

🌐 3. Connecting to ChatGPT Web (Chrome)

For browser-based ChatGPT (Web):

1. Add a Custom Connector/MCP tool in your ChatGPT Web interface.
2. Select SSE transport type.
3. Use the URL:

   https://flow.chatbulky.com/sse
   

4. Set Authentication to None (Disable OAuth).

---

🛠️ 4. Connecting Locally via Docker (Claude Desktop / Cursor)

If you are running the server locally via Docker, you can connect your Claude Desktop or Cursor client directly to the container using the mcp configuration file.

Add the following to your claude_desktop_config.json (or Cursor MCP config):

{
    "mcpServers": {
        "flow-agent": {
            "command": "docker",
            "args": [
                "exec",
                "-i",
                "flow-agent-server",
                "python3",
                "-u",
                "/app/flow_mcp_server.py"
            ]
        }
    }
}