MCP server

One server, any assistant

The Model Context Protocol is an open standard that lets an AI assistant use a product's tools with your permission. Handled Locator runs an MCP server, which is what powers the Claude and ChatGPT connections, and the same server works from Claude Code, Cursor, or any other MCP client.

• Endpoint: https://app.handledlocal.com/api/mcp
• Transport: Streamable HTTP
• Auth: sign in with your account (OAuth), or an API key
• Registry: published as com.handledlocal/locator in the official MCP Registry

It is included on every plan, including the free trial.

Developer clients

Claude and ChatGPT need no key: you sign in. Developer tools authenticate with an API key instead, created in the dashboard under Settings, then API & developer tools. Give the key write access if the assistant should change anything; a read-only key can look but not touch.

claude mcp add --transport http locator https://app.handledlocal.com/api/mcp \
--header "Authorization: Bearer hl_live_your_key"

{
"mcpServers": {
  "locator": {
    "url": "https://app.handledlocal.com/api/mcp",
    "headers": { "Authorization": "Bearer hl_live_your_key" }
  }
}
}

The tools

Seventeen, in five groups. Your assistant picks the right one from what you say.

• Locations: list, search, read, create, update and remove locations. Plain addresses are geocoded automatically, and distance questions are answered the way the widget answers them for visitors.
• Tags and buttons: list tags, tag or untag many locations at once, style a tag into a grouped, coloured filter with an icon, and put the same booking button, link or labelled fact on many locations in one step.
• Appearance and wording: read every setting with its current value, propose a change and see the exact differences, apply it after agreement, and restore any earlier version.
• Spotlight: rules that change what a visitor sees by where or when they search. A banner for a region or a date range, featured or exclusive locations, hidden ones. One tool, with an action to say what you want done.
• Going live: the embed code and your public site id, optionally locked to some tags for a single page.

The full tool table, with example prompts that have each been run against a live locator, is on the MCP server reference.

Not tools, and set from the dashboard instead: webhooks, the sync sources (Google Sheets, Airtable, Webflow CMS, HubSpot), Zapier, and the JavaScript API that runs in your page. The server tells the assistant where each lives, so asking it "set up a webhook" gets you the right page rather than an invented tool.

Webhooks are absent on purpose rather than for lack of room. An endpoint receives every location and every lead your account will ever have, so a tool that creates one would turn any instruction an assistant happens to read, in a web page or a pasted document, into a possible route out to somebody else's URL. That is a decision for an owner looking at the dashboard. A sync source is absent for the same shape of reason: it needs a credential to another system of yours.

Safety model

Changes to the live widget are two-step: propose returns the exact differences and changes nothing, and apply only runs after you agree. A proposal is refused if the locator changed in the dashboard meanwhile, rather than overwriting anyone's work. Configuration versions are all kept and restorable, deletes are soft, a connection reaches exactly one account, and keys are rate limited at 120 reads and 30 writes a minute.