Scan any URL for AI agent-readability from inside Gemini CLI (and, after migration, the Antigravity CLI). It checks sites against the Vercel Agent Readability Spec, the llmstxt.org standard, and agent-protocol manifests (MCP server cards, A2A, agents.json, agent-permissions.json, UCP, x402, NLWeb).

This extension is a thin wrapper around the existing agent-ready-mcp server and the agent-ready CLI — both published to npm and fetched on demand via npx. There is nothing to build.

It gives Gemini two ways to use Agent Ready:

• MCP tools the agent can call autonomously: scan_site, get_scan, ask.
• Slash commands you invoke directly: /agent-ready:scan, :get, :list, :ask.

• Gemini CLI installed.
• Node.js ≥ 20 (for the npx-launched MCP server and CLI).
• A Pro API key from https://agent-ready.dev/dashboard/api-keys for scan/get/list. The ask tool/command is public and needs no key.

gemini extensions install https://github.com/mlava/agent-ready-gemini

Local development (symlink instead of clone):

gemini extensions link /path/to/agent-ready-gemini

Manage it with the usual commands:

gemini extensions list
gemini extensions disable agent-ready
gemini extensions uninstall agent-ready

gemini extensions config agent-ready

This stores AGENT_READY_API_KEY securely (system keychain) and passes it to the MCP server. The optional AGENT_READY_API_URL overrides the API base URL (defaults to https://agent-ready.dev).

The slash commands shell out to the agent-ready CLI, which reads AGENT_READY_API_KEY from the environment. If gemini extensions config does not propagate the key into shell-command execution in your version, also export AGENT_READY_API_KEY=ar_live_... in your shell. The ask command never needs a key.

/agent-ready:scan https://example.com
/agent-ready:scan https://example.com --page-limit 25
/agent-ready:get V1StGXR8_Z
/agent-ready:list
/agent-ready:list --limit 5
/agent-ready:ask how is the score calculated?
/agent-ready:ask what does check S4 do? --type checks

Each command runs the corresponding agent-ready CLI subcommand and asks Gemini to summarize the JSON output. Gemini will prompt for confirmation before running the shell command.

Because the MCP server is registered, you can also just ask in natural language:

Scan example.com for AI agent-readability and tell me what to fix first.

Gemini will call the scan_site tool itself, then get_scan if the scan is still running.

Gemini CLI is being replaced by the Antigravity CLI (agy) for individual-tier users (deadline June 18, 2026). Import this extension as a native Antigravity plugin with:

agy plugin import gemini

This works cleanly here because the extension stays within the portable subset:

• The MCP server is local stdio (command/args), which carries over unchanged — the url → serverUrl rename only affects remote HTTP MCP servers, and this extension has none.
• commands/*.toml are collapsed into Antigravity skills automatically during import.
• GEMINI.md keeps working as the context file.
• No custom themes or Node-only API dependencies (the only things that don't migrate).

Invocation differs from Gemini CLI — this is expected:

• There is no /agent-ready:scan slash command in agy. The commands become skills (named scan, get, list, ask under ~/.gemini/config/plugins/agent-ready/skills/) that the agent invokes, and the MCP tools (scan_site, get_scan, ask) are called by the agent directly. Just ask in natural language, e.g. "use agent-ready to scan https://example.com" or "ask agent-ready how the score is calculated" (the latter is public, no key).
• API key: the imported ~/.gemini/config/plugins/agent-ready/mcp_config.json has env: null. For keyed scan/get/list, add your key there ("env": { "AGENT_READY_API_KEY": "ar_live_..." }) for the MCP tools, and/or export AGENT_READY_API_KEY=ar_live_... in your shell for the skill shell-outs. ask needs no key.

gemini-extension.json # manifest: MCP server + settings + contextFileName
GEMINI.md # context auto-loaded into the model
commands/agent-ready/ # /agent-ready:scan|get|list|ask (shell out to the CLI)

• Agent Ready: https://agent-ready.dev
• MCP server: https://www.npmjs.com/package/agent-ready-mcp
• CLI: https://www.npmjs.com/package/agent-ready-scanner

MIT © Agent Ready