Reduck

Reduck is a browser-automation layer for agents: an agent learns how a site works once, then stores that knowledge as a deterministic script instead of clicking through the page on every run and burning tokens.

Scripts are callable through Reduck MCP from Claude, Codex, and other MCP clients, or as an n8n node. They default to your own signed-in Chrome via a paired extension (no extra OAuth setup), or can run on Reduck's managed cloud browser for 24/7 public-page jobs.

Highlights:

• Separates "learn the site" from "run the script" — marketed as faster, less hallucinated, and cheaper in tokens;
• One integration for sites with no API or painful API access;
• A script library plus the option for agents to author private scripts;
• Site: https://reduck.ai ; docs: https://docs.reduck.ai .

Why it matters

Letting an agent drive a browser raw is slow, expensive, and brittle when the DOM shifts. Reduck freezes site knowledge into a script contract (inputs become node fields, outputs become the next item) — closer to "any website as a tool" than "hope the model finds the button." The n8n node lets non-coding teams reuse the same scripts in workflows.

Limits

A paired browser needs Chrome open on your machine. Managed browsers currently fit public pages better (login/cookie reuse is still evolving). Automating a third-party site must follow its terms; account security stays on you.

1. Create an account at https://reduck.ai , install the Chrome extension, and pair it;
2. Connect Reduck MCP in Claude/Codex (or similar), or install the n8n community node @reduck-ai/n8n-nodes-reduck with an API key;
3. Pick a library script or have an agent author a private one for your target site;
4. Choose paired Chrome (your logged-in sessions) or a managed browser (public pages, 24/7).