Antigravity CLI: install agy, sign in, skills, plugins and MCP

What is Antigravity CLI?

Antigravity CLI is Google's terminal coding agent, run with the agy command. You install one binary, sign in with a Google account, and give it tasks in a full-screen terminal interface where it plans, edits files, runs commands and shows you diffs to approve. Google's overview says every Antigravity surface, including the desktop app, runs on a shared agent harness.

It matters in 2026 because Google moved a large group of Gemini CLI users onto it. The Gemini CLI README now opens with: "Unpaid tier and Google One users: Gemini CLI was replaced by Antigravity CLI on June 18th, 2026." If you were in that group, this page covers the setup you need, and the Gemini CLI guide explains who can still use the older tool.

Everything below comes from Google's Antigravity docs as of 24 October 2026.

How do you install Antigravity CLI?

Google ships an install script per platform. On macOS, Linux and Googlebook:

curl -fsSL https://antigravity.google/cli/install.sh | bash

On Windows, use PowerShell or CMD:

# PowerShell
irm https://antigravity.google/cli/install.ps1 | iex

# CMD
curl -fsSL https://antigravity.google/cli/install.cmd -o install.cmd && install.cmd && del install.cmd

The binary lands in ~/.local/bin/agy on macOS and Linux, and in C:\Users\<username>\AppData\Local\agy\bin on Windows. Check it worked:

agy --version

The installer edits your shell profile to add agy to your PATH and to clean up old aliases. If you manage your profile yourself, the authentication page lists two flags for the install script: --skip-path and --skip-aliases.

The first time you run agy, it asks three things: a color scheme, a rendering mode (Alt-Screen, which takes over the terminal, or Inline, which prints into your scrollback), and whether you trust the current folder before it starts indexing files.

How do you sign in, and what do the plans include?

Run agy. If the CLI finds a saved session in your operating system keyring (Apple Keychain, Linux Secret Service or Windows Credential Manager), it signs in silently. Otherwise it opens your browser, where you pick a personal Google account or choose Business account for Gemini Enterprise.

Over SSH, the CLI cannot open a browser, so it prints a sign-in URL. Open it on your own machine, sign in, and paste the code it shows back into the terminal. To sign out and clear the keyring entry, type /logout in the prompt box.

There are two other ways in, both useful for CI or headless machines:

• Gemini API key. Set "modelProvider": "gemini" in ~/.gemini/antigravity-cli/settings.json, then export GEMINI_API_KEY="your-api-key". Setting only the variable does nothing, and the CLI reads the key only from the environment, not from .env files.
• Application Default Credentials. Run gcloud auth application-default login --project <your-project>, then export AGY_ADC_AUTH=true.

What you get depends on your Google AI plan. Google's plans page describes it like this:

Every plan includes Gemini models, including Gemini 3.8 Flash, as the core agent model, plus "Access to all product features, such as scheduled tasks and the CLI." A banner on the same page says Gemini 4 Argon is available for Google AI Pro and Ultra.

Pro and Ultra users can buy AI credits for usage above the quota. In the CLI, the useG1Credits setting (off by default) decides whether those credits are spent once your plan quota runs out. /usage shows model quota use and /credits shows your remaining credits. The plans page also says there is no bring-your-own-key option for extra rate limits.

One rule to know: Google's FAQ says using third-party tools to access Antigravity with your Antigravity login violates its terms, and that wrapping agy -p inside other coding agents is not supported on consumer plans.

How do you use the agy command?

Open a project folder and start a session:

cd my-project
agy                              # interactive terminal interface
agy "add unit tests for the auth handlers"   # start with a prompt
agy -c                           # continue the most recent conversation
agy -p "summarize this repo" --output-format json   # headless, for scripts

Inside a session, the agent indexes the workspace, writes an implementation plan and proposes changes. Press Ctrl+R to open the Artifact Review screen, step through diffs, and press Y to approve. Shift+Tab cycles execution modes: default, accept-edits, auto (in preview, where available) and plan. You can also set the mode at launch with --mode accept-edits, --mode auto or --mode plan.

The slash commands worth learning first, from the CLI reference:

Settings live in ~/.gemini/antigravity-cli/settings.json. The toolPermission key sets the default approval behavior: request-review (the default, which asks before write, bash, web and MCP tools unless an allow rule covers them), proceed-in-sandbox, always-proceed or strict (asks before every tool action, including file reads).

How do skills and rules work in Antigravity CLI?

Antigravity follows the open Agent Skills format: a folder with a SKILL.md file, YAML frontmatter, and optional scripts/, examples/ and resources/ folders. In Antigravity, description is required and name is optional, defaulting to the folder name. The agent sees each skill's name and description at the start of a conversation and reads the full file only when the task matches.

The skills page lists three places the CLI loads skills from:

• <workspace-root>/.agents/skills/<skill-folder>/ for one project
• ~/.gemini/antigravity-cli/skills/<skill-folder>/ for every project
• ~/.gemini/antigravity-cli/plugins/<name>/skills/ for skills that come with plugins

Every skill becomes a slash command automatically. Create deploy-staging and you can type /deploy-staging. The older .agent/skills folder still works for backward compatibility.

Rules are the always-on side. The CLI reads AGENTS.md and GEMINI.md at the repository root and in subfolders, plus single-topic files in .agents/rules/*.md. Those rule files need frontmatter with a trigger of always_on, model_decision, glob or manual, or they are silently dropped. Global rules go in ~/.gemini/AGENTS.md, ~/.gemini/GEMINI.md, ~/.gemini/config/rules/*.md or ~/.gemini/antigravity-cli/rules/*.md.

Two limits from the rules page: a single rule file is cut off after 24,000 bytes, and all always-on and global rules share a 20,000-token budget. Past that, the largest files become short pointers the agent reads on demand. Google's own advice is to write a rule for constraints and a skill for multi-step workflows.

How do plugins and MCP servers work?

A plugin bundles skills, subagents, rules, MCP servers and hooks behind one plugin.json manifest. In the CLI, name is required in that manifest. You manage plugins from the shell:

agy plugin install <plugin-name>                 # official marketplace
agy plugin install https://github.com/<owner>/<repo>
agy plugin install </path/to/local/plugin>
agy plugin list --json
agy plugin disable <plugin-name>
agy plugin uninstall <plugin-name>

A bare plugin name defaults to the antigravity-plugins-official marketplace. Installed plugins live in ~/.gemini/antigravity-cli/plugins/<plugin_name>/. Inside a session, /plugin opens the same manager with Discover and Installed tabs.

MCP servers are defined in their own JSON files rather than in settings:

• Global: ~/.gemini/config/mcp_config.json
• Workspace: <workspace>/.agents/mcp_config.json

{
  "mcpServers": {
    "sqlite-explorer": {
      "command": "node",
      "args": ["/usr/local/bin/sqlite-mcp-server.js"]
    },
    "my-remote-server": {
      "serverUrl": "https://api.example.com/mcp/",
      "headers": { "Authorization": "Bearer YOUR_API_TOKEN" }
    }
  }
}

Use command for a local stdio server and serverUrl for a remote one. The MCP page says remote servers must use Streamable HTTP; the legacy HTTP+SSE transport and the httpUrl key are not supported. Defaults are 5 minutes to connect and 3 minutes per tool call, which timeoutSeconds overrides. MCP tools ask for approval by default until you allow them with rules like mcp(server/tool) or mcp(server/*).

How do you move from Gemini CLI to Antigravity CLI?

Google's migration guide covers four changes.

1. First launch. When you run agy on a machine with Gemini CLI config, it offers a checklist of extensions and settings to convert, and moves your session tokens into the keyring.
2. Extensions become plugins. Run agy plugin import gemini. It converts extension manifests, turns legacy commands into skills and moves MCP definitions into mcp_config.json.
3. Skill folders move. Workspace skills go from .gemini/skills/ to .agents/skills/, and global skills from ~/.gemini/skills/ to ~/.gemini/antigravity-cli/skills/. The guide says you must move workspace skills yourself.
4. MCP config moves. Servers leave ~/.gemini/settings.json for the mcp_config.json files above, and url or httpUrl becomes serverUrl.

Context files need no changes. The guide says the agent "continues to parse and enforce rule constraints defined inside your active directory’s GEMINI.md and AGENTS.md files," and still reads ~/.gemini/GEMINI.md. The GEMINI.md guide covers writing that file, and the Gemini CLI skills guide covers the skill format both tools share.

How do you keep the same skills across Antigravity and other agents?

Because Antigravity uses .agents/skills/, the same folder works in other tools that read the Agent Skills format, such as Codex and Gemini CLI. See Codex skills and Claude Code skills for where each tool looks.

If you keep skills and prompts in more than one place, Kipt can hold them as Markdown in a private GitHub repo you own. You can import skills from a GitHub link, a folder or a .zip, then copy any skill folder into .agents/skills/ for Antigravity. To draft a new one, the free SKILL.md generator turns a prompt into a valid SKILL.md in your browser.