How to Migrate from Gemini CLI to Antigravity CLI

If Gemini CLI recently stopped accepting your individual Google account, reinstalling it will not solve the problem. Google ended Gemini CLI access for free, Google AI Pro, and Google AI Ultra individual accounts on June 18, 2026. For those accounts, Google’s supported terminal experience is now Antigravity CLI.

The good news is that you do not need to rebuild your agent setup from scratch. Antigravity CLI can detect an existing Gemini CLI profile during its first launch. Your GEMINI.md and AGENTS.md project instructions remain compatible, and the migration tools can convert extensions, settings, skills, and Model Context Protocol (MCP) servers.

Some paths and configuration formats have changed, though. Workspace skills move to .agents/skills/, Gemini extensions become Antigravity plugins, and MCP servers move out of settings.json into a dedicated mcp_config.json file.

This guide walks through the migration without deleting your existing Gemini CLI setup first.

Version note: AI developer tools change quickly, so compare commands and paths with the current official migration guide before proceeding.

Who needs to migrate?

The account you use with Gemini CLI determines whether the change affects you.

If you are using an individual account and see an authentication or account-access failure, move to Antigravity CLI rather than repeatedly signing out, reinstalling Gemini CLI, or regenerating local configuration.

Google’s Gemini Code Assist deprecation notice is the source of truth for affected account types.

What stays the same—and what changes

The migration is easier when you separate compatible project context from configuration that must move.

The most important point is simple: do not rewrite your GEMINI.md merely because the CLI changed. Antigravity uses the same workspace context rules and also reads AGENTS.md.

If you need help improving that file itself, see my existing guide to GEMINI.md examples and best practices.

Before migrating: inspect, do not delete

Keep the old Gemini configuration until Antigravity CLI is installed and verified. Start by checking which types of customization you actually use.

On macOS or Linux, inspect the relevant paths:

ls -la ~/.gemini
ls -la ~/.gemini/skills 2>/dev/null
ls -la .gemini/skills 2>/dev/null

Look for:

• A global or project-level GEMINI.md
• A project-level AGENTS.md
• Extensions or custom commands
• Global or workspace skills
• An mcpServers section in ~/.gemini/settings.json
• Custom themes, hooks, or experimental settings

Do not paste the contents of your complete Gemini directory into an AI chat or public issue. It may contain private project instructions, MCP server addresses, environment variables, or authentication-related data.

If you make a backup, keep it local and protect it like any other developer configuration. Never commit tokens, private headers, or copied credential files to Git.

Step 1: Install Antigravity CLI

Antigravity CLI runs natively on macOS, Linux, and Windows. The executable is named agy.

macOS and Linux

Run Google’s official installer:

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

The installer places the binary at ~/.local/bin/agy by default.

If your organization requires reviewing scripts before execution, download and inspect the installer from the same official domain before running it.

Windows PowerShell

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

Windows Command Prompt

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

Google’s current platform commands are maintained in the official installation and authentication guide.

Step 2: Start agy and complete the first-launch import

Launch the new CLI from a terminal:

agy

When Antigravity finds an existing Gemini CLI profile, its onboarding flow presents an interactive checklist. Choose the extensions and global configurations you want to convert.

According to the migration documentation, first-launch onboarding can:

• Convert selected Gemini extensions and global configuration
• Move an active session into the operating system’s secure keyring
• Map supported visual and rendering settings into the new profile

On a local computer, Antigravity attempts to authenticate through the operating system keyring. If it cannot find a valid session, it opens Google Sign-In in the default browser.

When running over SSH, agy prints an authorization URL. Open that URL locally, complete authentication, and paste the resulting authorization code back into the remote terminal.

Step 3: Confirm that GEMINI.md and AGENTS.md still load

You do not need to rename your existing project context files. Antigravity CLI reads:

• GEMINI.md and AGENTS.md in the active workspace
• Global instructions from ~/.gemini/GEMINI.md

Open a non-sensitive test project containing your context file, start agy, and ask the agent to summarize the project rules it has loaded. Compare the response against the actual file instead of assuming detection worked.

Then give it a small, reversible planning task that should trigger one unmistakable instruction. For example, if your context says to use a specific package manager or to remain read-only until approval, check whether the agent follows that rule.

This verifies behavior without letting the first migration test make consequential changes.

Step 4: Move workspace skills to the new path

Antigravity changed where it discovers skills.

Workspace skills require manual attention. If your repository contains .gemini/skills/, move each relevant skill into .agents/skills/ and review its contents before committing the change.

For a simple workspace with no conflicting .agents directory, the structure should look like this:

your-project/
├── AGENTS.md
├── GEMINI.md
└── .agents/
    └── skills/
        └── your-skill.md

After restarting agy in that workspace, a registered skill should appear as a slash command. Test the command with non-sensitive inputs before using it in production work.

Do not blindly move an entire directory if your repository already has an .agents structure. Merge deliberately so you do not overwrite instructions or skills used by another agent.

Step 5: Convert Gemini extensions into Antigravity plugins

Antigravity calls extensibility bundles plugins rather than extensions. To import locally installed Gemini extensions, run:

agy plugin import gemini

The importer searches legacy directories, reads extension manifests, and converts supported components into Antigravity’s plugin layout. It can process skills, agents, legacy commands, and MCP server definitions.

List the resulting plugins:

agy plugin list

Read the import output carefully. A successful command does not necessarily mean every component was present or converted. The tool reports components as processed or skipped, so investigate unexpected skips before assuming the migration is complete.

Antigravity stores imported plugins under:

~/.gemini/antigravity-cli/plugins/<plugin_name>/

The plugins and skills documentation explains the current manifest and directory structure.

Step 6: Move and update MCP server configuration

Gemini CLI could define MCP servers inside ~/.gemini/settings.json. Antigravity separates them into dedicated files:

# Global MCP servers
~/.gemini/config/mcp_config.json

# Project-specific MCP servers
.agents/mcp_config.json

The file contains one mcpServers object. A local stdio server can look like this:

{
  "mcpServers": {
    "local-tools": {
      "command": "node",
      "args": ["/absolute/path/to/server.js"]
    }
  }
}

For a remote MCP server, Antigravity expects serverUrl:

{
  "mcpServers": {
    "remote-tools": {
      "serverUrl": "https://example.com/mcp/"
    }
  }
}

If your legacy definition uses url or httpUrl, replace it with serverUrl. The old keys are not supported for remote SSE, Streamable HTTP, or WebSocket connections.

Open the interactive MCP manager from inside Antigravity CLI:

/mcp

Use it to check whether each server is active, disconnected, or still loading. Reload the configuration and inspect connection logs when a server does not start.

Be especially careful with authentication. Do not commit API keys or bearer tokens in a workspace mcp_config.json. Prefer the authentication mechanisms supported by the server and keep machine-specific secrets outside version control.

For a simple explanation of the protocol itself, read What is MCP in AI?. For the current Antigravity schema and authentication options, use Google’s MCP documentation.

Step 7: Verify the migration before cleaning anything up

Do not judge success only by whether agy opens. Verify each layer you depend on.

Use this checklist:

• agy launches from a new terminal session
• Authentication completes without repeatedly reopening the browser
• The agent can summarize the intended GEMINI.md or AGENTS.md rules
• A small task follows one distinctive project instruction
• Workspace skills appear as slash commands
• agy plugin list shows the expected converted plugins
• /mcp shows the expected servers and connection states
• One harmless MCP read operation works with approval
• No secrets or private configuration were added to Git
• A reversible test change produces a reviewable diff

Keep Gemini CLI configuration in place until these checks pass. A successful migration is one you can verify, not one that merely completed without an error message.

Common migration problems

agy: command not found

On macOS and Linux, the installer normally places agy in ~/.local/bin. If a new terminal still cannot find it, check the file and your current PATH:

ls -l ~/.local/bin/agy
echo "$PATH"

If necessary, add the directory to your shell configuration:

export PATH="$HOME/.local/bin:$PATH"

Reload the appropriate configuration—for example, source ~/.zshrc for Zsh—or open a new terminal.

The CLI repeatedly asks you to sign in

Antigravity stores credentials in the operating system’s secure keyring. Repeated authentication can indicate that the keyring is locked or unavailable, particularly in a headless Linux or SSH environment.

Check your system keyring permissions and the official troubleshooting guide. Inside the CLI, /logout clears the saved Antigravity authentication profile so you can start a fresh sign-in.

Do not delete unrelated system credentials while troubleshooting.

GEMINI.md works, but a skill is missing

Context files and skills use different discovery rules. Leave GEMINI.md where it is, but move workspace skills from .gemini/skills/ to .agents/skills/. Restart agy from the project root and check for the generated slash command.

An MCP server no longer connects

Check three migration-specific causes first:

1. The server definition is still nested inside the old settings.json.
2. A remote server still uses url or httpUrl instead of serverUrl.
3. The configuration was placed globally when it should be workspace-specific, or vice versa.

Open /mcp to inspect the server state and logs before changing multiple variables at once.

A Gemini extension was not converted

Run agy plugin import gemini, then read the component-level output. If the importer reports that a component was skipped, compare the old extension structure with the current Antigravity plugin format rather than repeatedly rerunning the same command.

Should you uninstall Gemini CLI afterward?

There is no benefit to rushing this step. First verify your context files, skills, plugins, MCP servers, and authentication in Antigravity CLI. Keep a protected local backup of configuration you may need to reference.

Individual Google Sign-In access ending does not make every Gemini CLI file useless. Your GEMINI.md, project rules, and reusable workflow documentation remain valuable—and some enterprise or API-key workflows continue to use Gemini CLI.

Only remove the old executable or configuration after you know which files Antigravity still reads and which migrated copies are now authoritative.

The migration is mostly about configuration, not rewriting your workflow

For most individual users, the move from Gemini CLI to Antigravity CLI has four practical parts:

1. Install and authenticate agy.
2. Keep GEMINI.md and AGENTS.md in place.
3. relocate skills and import extensions as plugins.
4. Move MCP servers into mcp_config.json and update remote URL keys.

The parts that describe your project—its goals, constraints, conventions, and non-negotiable rules—should survive the tool transition. That is one reason I treat project context as durable documentation rather than a prompt written for a single model.

If you want to explore that idea further, read What is context engineering in AI? and my practical GEMINI.md guide.

Official sources

• Google: Gemini Code Assist consumer-account deprecation
• Google: Migrating from Gemini CLI
• Google: Antigravity CLI installation and authentication
• Google: Antigravity plugins and skills
• Google: Antigravity MCP configuration
• Google: Antigravity CLI troubleshooting