Give Claude a Working Browser
Something you can use today.
Give Claude a Working Browser
The Weekly Build
Most of what Claude can't do for you comes down to one thing. It can't see the web.
Ask it to check a price, pull a spec sheet or scrape a competitor's pricing page and it'll either refuse or make something up. That's a training cut-off problem, and every frontier model has one, with the gap between the cut-off and today only growing.
Firecrawl's MCP server fixes it. It gives Claude an actual scraping and search tool it can call whenever it needs to look at a live page, rather than guess from memory. Twenty minutes, no coding beyond copying a config file, and here's how.
1. Get a Firecrawl API key. Sign up at firecrawl.dev, and the free tier gives you enough credits to test this properly.
2. Open Claude Desktop's config file. On a Mac that's `~/Library/Application Support/Claude/claude_desktop_config.json`. On Windows it's in `%APPDATA%\Claude\`. If the file doesn't exist, create it.
3. Add the Firecrawl server, pasting this in with your own key:
{
"mcpServers": {
"firecrawl": {
"command": "npx",
"args": ["-y", "firecrawl-mcp"],
"env": {
"FIRECRAWL_API_KEY": "your-key-here"
}
}
}
}4. Restart Claude Desktop, and fully quit it rather than just closing the window.
5. Check the connection. Look for a small hammer icon near the message box, click it, and Firecrawl should be listed as an available tool.
6. Ask it to fetch something real. Try: "Go to (any public pricing page) and pull out every plan name and price into a table," then watch it call the tool rather than guess.
7. Push it further, asking it to scrape three competitor pages and compare them. This is where it earns its keep.
What you end up with is a Claude that reads the live web on request instead of working from a memory that's already months stale. Firecrawl's own server has passed 7,000 stars on GitHub, which tells you plenty of people are running this exact setup daily rather than trying it once and forgetting about it.
One warning, and it gets its own sentence because it matters: scraping sites you don't own can breach their terms of service. Stick to your own pages and public data until you've checked.
This Week's Challenge
Open Claude Desktop with Firecrawl connected and ask it to scrape your own homepage, listing every claim you make about turnaround time or pricing. Fifteen minutes gets you a plain list of what your own site promises a customer, in writing, whether you meant it to or not.
Worth knowing before someone else finds it first.
Try it and tell me in the comments what happened. I read every one.
Forward this to one person who should be using AI better than they are. Reply with what you built, tried or broke this week. I read every one.
Gareth, founder of The Anthropic Stack (theanthropicstack.com)