What it is
The Browser Automation API takes a URL, opens it in a real browser so the page's JavaScript runs, and hands back the title, the visible text and the links on the page. It's delivered over MCP into Claude or another MCP-capable assistant — there's nothing to install and no browser running on your own machine.
The reason it exists is a specific gap: a plain HTTP fetch only downloads the HTML a server sends and stops there. Most modern sites build their actual navigation and body afterward, in the browser. So a fetch often comes back as a shell with almost nothing readable in it. On one example page run through the tool, the full rendered version returned 4,338 characters of text and 100 links, compared against the free preview of that same render, which returned only the title and the first 500 characters, with links withheld.
You can try it without an account or a card. Connect an MCP-capable assistant to the product's MCP endpoint and call `browser_fetch`. The free preview returns the real page title and the first 500 characters of the real render — links are withheld at this tier — and is capped at 10 renders per day, so you can confirm the tool actually works on pages you care about before committing to a plan.
Paid plans raise the daily cap and add more of the page. Starter, at $29/month, gives 500 page renders per day, full page text up to 20,000 characters, and every link on the page with its anchor text. It uses the same API key as any other Kcalbin MCP product you subscribe to. Pro, at $99/month, raises the cap to 2,500 renders per day with the same feature set as Starter. Both are daily quotas, not unlimited access.
What it does not do
The product page is direct about where this tool stops, and it's worth reading before you buy:
It does not rotate proxies or use residential IPs. Every request comes from one datacenter address, identified as `KcalbinBrowserAPI/1.0`. It does not bypass anti-bot protection — a site running commercial bot defenses will block it, and there's no attempt made to get around that. It does not let you choose the country a request appears to come from; there is no geotargeting. And it is read-only: it renders a page and extracts from it, but it does not click buttons, fill out forms, log in, or take screenshots.
It also only works on public pages. Private, internal and cloud-metadata addresses are refused outright, as is any hostname that resolves to one.
None of this is a defect to be fixed later — it's a boundary the product states plainly. If what you need is rotating residential proxies, anti-bot bypass, or renders in the millions per month, that's a dedicated scraping platform's job, and the page says so without hedging. This tool is aimed at a narrower case: reading a page properly from inside the assistant you're already using, billed through the same key as your other Kcalbin MCP tools.
It's also worth being clear about what the free preview itself doesn't promise. The 500-character preview confirms that the tool can render your page and pull a real title and text out of it — it doesn't tell you what the full rendered output will contain, since links are withheld and the text is cut short. The 10-renders-a-day cap is a real daily limit, not a rough guideline, and it reflects the capacity of the hardware serving it rather than a marketing number. The same is true of the paid quotas: 500 and 2,500 renders per day are described as real limits, not just suggested usage.
The practical read
For a business evaluating whether JavaScript-rendered pages are usable in an AI workflow, the free tier is a reasonable first step: point it at a page you actually care about, see the title and the opening text come back, and judge from that whether the rendering is doing what you need. It won't tell you how the tool handles bot protection on a different site, or what a full 20,000-character extract looks like — for that you'd need a paid plan and a test on the pages that matter to you. What the preview does show, honestly, is whether the basic mechanism — open the page, let it render, read what's there — works on your own content before any money changes hands.