Connect Gavana to Antigravity

Antigravity is Google’s agent-first IDE. MCP works the same way across all three surfaces — Antigravity 2.0, the Antigravity IDE, and the Antigravity CLI — because they share one configuration format and one config file.

Antigravity is an IDE process, not a browser tab, so it reaches Gavana’s hosted endpoints directly with no origin restrictions.

Connect with full access

https://app.gavana.ai/mcp

This is the default Gavana connection. It requests all eight permissions and remains active until you disconnect or revoke it.

Where the config lives

~/.gemini/config/mcp_config.json      global — all Antigravity surfaces
<project>/.agents/mcp_config.json     per-workspace — this project only

Add the server

Normal connection:

{
  "mcpServers": {
    "gavana": {
      "serverUrl": "https://app.gavana.ai/mcp"
    }
  }
}

That bare entry is the complete configuration. Antigravity handles OAuth automatically for servers that support dynamic client registration, and Gavana does — so no oauth block, no client ID, no client secret.

If the file already has an mcpServers object, add the gavana entry to it rather than replacing the whole file.

Getting to the config from the UI

Each surface has its own route to the same file.

Antigravity 2.0 — click the Settings button at the bottom left of your screen, select Customizations, and review the Installed MCP Servers section. Add MCP opens the MCP Store; each installed entry gets a trash-can to uninstall, a toggle to disable or enable, and a refresh button.

Antigravity IDE — click … at the top of the editor’s agent side panel and select MCP Servers, then Manage MCP Servers, then View raw config. That opens mcp_config.json for direct editing. Once installed, the server’s tools and resources are automatically available to the editor.

Antigravity CLI — type /mcp in the prompt panel and press Enter to open the interactive MCP Manager Overlay. It shows live status rings for active, disconnected, and loading servers, and lets you reload server configurations or inspect real-time connection logs.

Complete the OAuth flow

This is the step people get stuck on. Antigravity’s OAuth is not a pure browser round-trip — you have to bring a code back into the app by hand.

The server reconnects automatically once the code is accepted. Access tokens are stored at ~/.gemini/antigravity/mcp_oauth_tokens.json; expired tokens are refreshed for you and invalid ones are removed.

Verify

In the CLI, /mcp shows gavana with a live status ring. In 2.0 and the IDE, the server appears under Settings → Customizations → Installed MCP Servers with a toggle and a refresh button.

Then ask for something read-only:

Using Gavana, list my canvases and read the most recently updated one. Report its
handle, revision, node count, and connection count. Do not change anything.

Tool approvals

You can widen that with policy patterns — mcp(gavana/canvas_list) for a single tool, mcp(gavana/*) for the whole server, mcp(*) for everything. If you auto-approve anything, auto-approve individual read tools, never the server wildcard on the full endpoint.

Resulting permission surface

Read-only endpoint

Tools exposed: guide_search, guide_get, canvas_list, canvas_get, canvas_validate, get_canvas_image, open_canvas, element_collection_list, element_get, element_history, and element_list — 11 in total. It issues canvas:read and element:read.

find_video_models is absent even though it only reads. It requires video:generate, and this endpoint never issues that scope.

Cost exposure: none.

Full endpoint

Sixteen tools: everything above, plus canvas writes, workflow authoring, and generation. Three of them can charge your connected AI provider — run_canvas_workflow, generate_image_in_canvas, and generate_video. Full schemas: hosted MCP tools.

Cost exposure: real.

disabledTools on the server entry lets you hide specific tools from the agent, and disabled switches the whole server off without deleting it. Both are convenience, not a security boundary — the endpoint and the Ask-mode approvals are the boundary.

The local stdio server

Antigravity can also launch Gavana’s local server as a child process, which exposes 57 tools instead of 29 and authenticates with a Personal Access Token rather than OAuth. Use command, args, and env in place of serverUrl:

{
  "mcpServers": {
    "gavana": {
      "command": "npx",
      "args": ["-y", "@gavana.ai/mcp@0.2.0"],
      "env": {
        "GAVANA_BASE_URL": "https://app.gavana.ai",
        "GAVANA_AGENT_TOKEN": "PASTE_YOUR_TOKEN_HERE"
      }
    }
  }
}

Needs Node.js 20 or newer. The real permission surface here is the token’s scopes, not the tool list — a tool called outside them fails with a permission error. Add "GAVANA_MCP_READ_ONLY": "true" for a hard read-only server, or "GAVANA_MCP_TOOLSETS" to narrow the catalog; scoping the token is still the stronger control.

Disconnect

Remove the gavana entry from mcp_config.json, or uninstall it from Settings → Customizations → Installed MCP Servers. Then revoke the OAuth delegation from Gavana’s Personal Access Tokens screen — revocation takes effect on the next request. Deleting ~/.gemini/antigravity/mcp_oauth_tokens.json clears the local copy but does not revoke anything server-side.

If you used the local stdio path, revoke the Personal Access Token as well.

Troubleshooting

More: MCP troubleshooting.

Security notes

• The hosted endpoints accept OAuth bearer tokens only. Do not attempt to use a cba_… Personal Access Token there — that is the local MCP and API path.
• Webhooks are not available through MCP by design: a signing secret must never enter model-visible tool arguments or tool logs. Use the CLI or the API when you need callbacks.
• Generation charges the AI provider connection on your Gavana account, not Google.

Related

• Connection overview
• Cursor
• Other clients
• The safety contract
• MCP authentication
• Hosted MCP tools
• Local MCP tools
• Create an Agent Access Token