Antigravity

Google Antigravity can use Bifrost through a manual MCP server entry. Antigravity’s visible Add MCP flow is a curated marketplace, but the app also reads local MCP configuration from ~/.gemini/config/mcp_config.json.

For Antigravity’s underlying host conventions, see the official MCP and Skills documentation.

Configure MCP

Build Bifrost first:

cargo build --bin bifrost

Add a bifrost entry to ~/.gemini/config/mcp_config.json:

{
  "mcpServers": {
    "bifrost": {
      "command": "/path/to/bifrost/target/debug/bifrost",
      "args": [
        "--root",
        "/path/to/workspace",
        "--mcp",
        "symbol|extended"
      ]
    }
  }
}

Restart Antigravity or open Settings -> Customizations and click Refresh. The Installed MCP Servers section should show bifrost with the Bifrost tools enabled.

Add Skills

Antigravity documents skills as folders with SKILL.md under either the workspace or global skill directory:

• <workspace-root>/.agents/skills/<skill-folder>/
• ~/.gemini/antigravity/skills/<skill-folder>/

In Antigravity 2.2.1 validation, workspace-local skills loaded reliably and appeared in project-specific settings. If global skills do not appear in your app session, install Bifrost’s generic code-intelligence skills into each target workspace:

bifrost --root /path/to/workspace --install-skills --target project --mode copy

Use --skills-root ~/.gemini/antigravity/skills only when you explicitly want to install into Antigravity’s global app-state root. See CLI for the full option list.

Then restart Antigravity. Open the project-specific settings page, not only global Customizations. The project Customizations section should list bifrost-code-navigation, bifrost-code-reading, and bifrost-codebase-search alongside any global skills.

Use It for Guided Review

Bifrost’s default generic skills expose analyzer-backed navigation, reading, search, and usage guidance. Use them as the code-intelligence layer for guided review workflows: ask Antigravity to load the relevant Bifrost skill, inspect the changed files, use Bifrost MCP tools for source context, and present review findings with file and line references.

Validate the Setup

Use a source-backed prompt that forces an MCP tool call:

Use the bifrost-code-reading skill. Inspect the current changes, use the Bifrost MCP get_summaries tool on src/analyzer/usages for source context, and report review findings with file and line references.

Antigravity should ask for MCP permission the first time it calls the tool. A successful smoke should show a bifrost / get_summaries tool call before it presents review context or findings.

Avoid prompts that only ask about README.md or docs files; those can pass through ordinary file reading without proving the MCP server ran.

Can My Agent Run RQL?

The configuration above uses symbol|extended. In a fresh Antigravity session, confirm that the enabled Bifrost tool list includes query_code, then call it with the inline JSON fields {"match":{"kind":"declaration"},"limit":1}. To validate saved RQL, check a workspace file named bifrost-smoke.rql containing (limit 1 (declaration)), then call query_code with {"query_file":"bifrost-smoke.rql"}.

The inline call is canonical JSON. MCP accepts RQL only from a workspace .rql file through query_file. The separately installed skills provide guidance but do not expose MCP tools themselves. See MCP query and RQL availability for the full surface matrix and Agent Result Safety before making completeness claims.