Grabbit CLI

Turn browser actions into reusable API workflows. Browse a site, capture what you do, and Grabbit generates stable cURL-ready workflows for automation, data extraction, and integrations.

Install

npm i -g @cole-labs/grabbit

Get started

1. Sign in (one-time)

grabbit auth

Your browser opens to pair. Sign in, then enter the code from the page into your terminal.

2. Open a site and interact

# Open a page (use --headed if you need to see the browser, e.g. for logins)
grabbit browse --session myrun open https://example.com

# Inspect and interact
grabbit browse --session myrun snapshot
grabbit browse --session myrun click @e3
grabbit browse --session myrun fill @e5 "search term"

3. Save and generate

grabbit save --session myrun "Describe what you want. Example: Extract product titles. Output: { titles: string[] }"

4. Check the result

grabbit check <task-id>

When it’s done, you’ll see inputs, outputs, and a cURL example.

Prompt tips

Be specific. Include real examples you saw on the page and the output shape you expect.

• Static data: "Extract blog titles, e.g. 'Cooking for Beginners'. Output: { titles: string[] }"
• Interactive flows: Describe each step with example data (name, address, etc.)
• Generalize: Even for one value, phrase it so the workflow works for many cases

Browser commands

Run grabbit browse --help for the full list. Common commands:

• Navigate: open, back, forward, reload, close
• Inspect: snapshot, get text, get html, get value, is visible
• Interact: click, fill, type, press, select, hover, focus
• Wait: wait <ms> or wait <selector>
• Debug: screenshot, console, errors

Use workflows in AI agents

To add Grabbit as a skill for AI agents (e.g. Cursor, Claude):

grabbit skill install

Troubleshooting

• Not authenticated: Run grabbit auth, then retry.
• No requests recorded: Reload the page, interact with it, then run save again.
• Always use --session <name> so captures stay separate.

Credits

Grabbit’s browser automation is powered by a fork of agent-browser, licensed under the Apache License 2.0.

License

MIT