Gemini MCP: How to Connect the Gemini App and Gemini CLI, and Make Real 3D Models

Where does Gemini support MCP?

The Model Context Protocol lets an assistant use tools from other services. Google added it to Gemini in two very different products, and the setup is different in each:

• The Gemini app (gemini.google.com and the mobile apps) calls them custom Connected Apps. You add a remote server by URL.
• Gemini CLI, Google's terminal agent, reads servers from a settings file and supports local servers that run on your machine as well as remote ones.

Google's Antigravity editor supports MCP too, and several engines can configure it for you: Roblox Studio's Quick connect lists Gemini CLI and Antigravity, and Unreal Engine's MCP plugin can write a Gemini config.

How do you add an MCP server to the Gemini app?

1. 1

   Open Connected Apps

   On gemini.google.com, open Settings and then Connected Apps. If you do not see it, look under Personal Intelligence.

2. 2

   Add a custom app

   Under Custom apps, click Add a custom app (or use the Add a custom app link field) and enter the MCP server URL.

3. 3

   Sign in to the service

   For 3D AI Studio, enter https://mcp.3daistudio.com/mcp and complete the sign-in page that opens. There is no API key.

4. 4

   Use it on web or mobile

   Custom apps you add on the web also work in the Gemini mobile app. Gemini asks you to confirm write actions before it runs them.

How do you add an MCP server to Gemini CLI?

One command adds a remote server. Gemini CLI finds 3D AI Studio's OAuth settings by itself, so the only other step is one sign-in command.

gemini mcp add --transport http 3daistudio https://mcp.3daistudio.com/mcp

# then, inside Gemini CLI:
/mcp auth 3daistudio

Or add it to the settings file by hand and restart Gemini CLI:

{
  "mcpServers": {
    "3daistudio": {
      "httpUrl": "https://mcp.3daistudio.com/mcp"
    }
  }
}

• httpUrl or url: older Gemini CLI versions use httpUrl for Streamable HTTP servers and url for SSE. Newer versions accept url with an optional "type": "http". If one form is rejected, try the other.
• OAuth needs a browser. Sign-in opens your browser and returns to a localhost address, so it does not work over SSH without forwarding or in a headless container.
• Restart after changes. Gemini CLI reads its settings at startup.

Can Gemini make 3D models?

Partly, and it is worth being precise. In April 2026 Google added interactive 3D visualisations to the Gemini app: choose the Pro model, ask it to "show me" or "help me visualise" something, and you can rotate a model or play with a simulation. That is great for understanding how something works. It is not a model you can download, print or put in a game.

If what you want is a file, the MCP route is the shortest. Gemini stays your interface, 3D AI Studio does the generation, and you get a textured model with UVs in standard formats.

How do you go from a Gemini image to a 3D model?

Gemini is very good at images thanks to Nano Banana, and a good image is the best start for a 3D model. Ask for a clean reference first, then make it 3D.

Each of these started as a single image and became a textured 3D model. Pick one and drag to rotate it, or switch to clay to see the mesh.

Choose an input image

Prompts that work in Gemini

Use 3D AI Studio to make a stylized treasure chest for my game and give me a GLB.

Make an image of a cute robot figurine, front view, plain white background.
Then use 3D AI Studio to turn it into a 3D model and give me an STL 10 cm tall.

Use 3D AI Studio to turn this photo of my dog into a 3D figurine.

Use 3D AI Studio to make a low poly version of my last model, under 5,000 triangles.

Gemini MCP not working? Common fixes

Frequently asked questions