# Antigravity

Source: /docs/mcp/connect/antigravity

Antigravity reads MCP servers from one global config file, `~/.gemini/config/mcp_config.json`, shared by the Antigravity IDE and the Antigravity CLI. Add OrbitOrca there and every workspace can use it, in both.

### Get a token [#get-a-token]

Open the dashboard's [Connect an agent](https://app.orbitorca.com/mcp) page and create a token. **Read & write** is the recommended choice — it lets the agent deploy, restart, and manage your apps hands-free. Pick read-only if the agent should only look. The full token is shown **once**, at creation — copy it now.

### Add the server [#add-the-server]

Add this block to `~/.gemini/config/mcp_config.json`, with your token in place of `{{token}}`. If the file already has an `mcpServers` object, add the `orbitorca` entry inside it:

```json
{
  "mcpServers": {
    "orbitorca": {
      "serverUrl": "https://api.orbitorca.com/mcp",
      "headers": {
        "Authorization": "Bearer {{token}}"
      }
    }
  }
}
```

In the Antigravity CLI, run `/mcp&#x60; to see the server. In the IDE you can open that file from the &#x2A;*…** menu at the top of the agent panel: **MCP Servers**, then **Manage MCP Servers**, then **View raw config**.

Remote servers use `serverUrl` here — Antigravity doesn't accept `url` or `httpUrl` for them.

### Try it [#try-it]

Ask the agent to &#x2A;"list my OrbitOrca apps"*. If it comes back with your apps, you're connected.

## Approvals [#approvals]

Antigravity asks before it runs an MCP tool by default: tools that no policy covers run in **Ask** mode, so you approve each call. You can allow tools in your policy settings — see [Antigravity's MCP docs](https://antigravity.google/docs/mcp). Keep the write tools on Ask unless you're comfortable with the agent deploying and changing apps without checking with you.

The dashboard's [Connect an agent](https://app.orbitorca.com/mcp) page generates this block with your token already filled in — copy it from there if you'd rather not edit by hand.

Next: [what a token can and can't do](/docs/mcp/scopes-and-safety).