Handsome Local
MCP

Use the gallery inside your AI assistant

Every published website here, 72 at the moment, can be searched from Claude Code, Claude.ai, Cursor and any other tool that speaks the Model Context Protocol. Free, no key, one line to install. Ask for 3 minimal electrician sites with the phone in the header and the screenshots come back in the chat.

Install

Claude Code. Run this once, then start a new session.

claude mcp add handsome-local --transport http https://handsomelocal.com/mcp

Claude.ai. Settings, Connectors, Add custom connector, and paste the address. Turn it on in any chat.

https://handsomelocal.com/mcp

Cursor, Codex, Windsurf and the rest. Most clients take the address on its own. For one that wants a config block:

{
  "mcpServers": {
    "handsome-local": {
      "url": "https://handsomelocal.com/mcp"
    }
  }
}

The endpoint speaks Streamable HTTP with no authentication. Older clients that still send stateless requests in the previous shape are accepted too.

What to ask

  • "Show me 3 plumber websites from the UK with a review count."
  • "5 dark roofing sites built on WordPress."
  • "Landscaper websites in the US that show a service area map."
  • "Which trade sites here use DM Sans?"
  • "Get the full review for cherry-blossom-plumbing."

Three tools sit behind it. search_websites takes a trade, region, style, colour, platform, a set of features, a typeface or free text and returns up to 8 matches, newest first. get_website returns one entry with the whole review. list_filters lists the trades and tags that have entries, with counts, so an assistant can see what is there before it searches.

What comes back

Each result is the business name, its trade and town, the first sentence of the review, the credit for whoever built it, the live site, the entry page on this site and a screenshot of the top of the page. Claude Code shows the screenshot inline; Claude.ai shows it through the image link. The same data is at https://handsomelocal.com/mcp/index.json as one JSON file, rebuilt with the site.

The screenshots belong to the businesses shown and travel with their credit, exactly as they do on the site. Nothing is offered as a file to keep. If a site of yours is here and you would rather it was not, say so through the contact form and it leaves the site and the endpoint at the next build, within 2 working days.

Limits and privacy

60 requests a minute per address, which is plenty for one person and enough to notice a script. Each tool call is logged with the filters used and the number of results, alongside the requesting address, in Cloudflare's Worker logs for up to 7 days. Nothing else from the conversation reaches this site. Details in the privacy page.

This runs on the same Cloudflare Worker as the site, from the same build, so what the assistant sees is what the site shows. Questions and broken bits go through the contact form.