For AI assistants

Give your assistant the store database

An MCP server that lets Claude, Cursor or any MCP client search 7,708,982 live stores across all 99 fields — and ask you first before it spends anything.

Install it

One block in your MCP client's config file. Nothing to download first.

{
  "mcpServers": {
    "myleadfox": {
      "command": "npx",
      "args": ["-y", "myleadfox-mcp"],
      "env": {
        "MYLEADFOX_API_KEY": "mlf_live_YOUR_KEY",
        "MYLEADFOX_MCP_MODE": "safe",
        "MYLEADFOX_SPEND_GATE": "500"
      }
    }
  }
}

Your key comes from the app, under Settings → API keys. In Claude Desktop the file is claude_desktop_config.json; Claude Code, Cursor and VS Code each have their own, and the block is the same in all of them.

It asks before it spends

The question everyone asks second, answered first: an assistant in a loop cannot quietly run up a bill.

Anything from 500 credits up stops

The first call returns a price and a one-time token, and never reaches our API. Your assistant shows you the price. Only a second call carrying that token goes ahead.

Tokens are single use, expire in five minutes, and are tied to the exact request that was priced. Change a filter after the preview and the token is worthless.

The price is the real number

Before spending, the server counts the match set for a single search credit. So a request for three pages against twelve matching stores is priced at twelve leads, not three hundred — because you are charged per row delivered, never per row requested.

A search that returns a count and no rows is the cheapest thing in the product, which is what makes honest pricing possible at all.

Three modes, and safe is the default

Mode	Tools	What it can do	When to use it
readonly	2	Counting and pricing only. Nothing that returns a row, nothing that can spend a lead.	A shared or unattended assistant, or anyone evaluating what the data holds before trusting it with credits.
safedefault	4	Adds rows and lookup, both behind the spend gate.	The default. An assistant can deliver real data, and cannot spend past the threshold without you seeing the price.
full	5	Adds contact enrichment.	Your own machine, your own key, work you are watching.

The mode lives in your config file, which the assistant cannot edit. A tool outside the current mode is refused when called, not merely hidden from the list.

The tools

Tool	Mode	Costs	What it does
store_search	readonly	1 search credit	Counts a filter set and returns a search id. Never returns rows, so it is how an assistant sizes a job before spending on it.
describe_filters	readonly	nothing	The filter kinds, operators and column names, with worked examples. No API call at all — an assistant can learn the surface for free instead of guessing and spending a credit on a refusal.
store_rows	safe	1 lead per row delivered	A page of up to 100 stores with every field. Above the spend threshold it returns a price and does nothing until you agree.
store_lookup	safe	1 lead per domain found	Fetch known stores by domain, up to 100 at a time. Domains we do not hold cost nothing.
store_enrich	full	1 enrichment credit per contact found	Contact details for stores. Only in full mode, because this pool is the most expensive and the result is personal data.

Common questions

What is an MCP server?

A small program that exposes a product to an AI assistant as a set of tools it can call. It runs on your own machine, holds your API key, and speaks to our API like any other client. Claude, Cursor, VS Code and a growing list of others can use one.

Can the assistant spend my credits without asking?

Not past the threshold. Anything that would spend 500 or more returns a price and a token and does NOT call the API. The assistant shows you the price; only a second call with that token proceeds. Tokens are single use, expire in five minutes, and are bound to the exact arguments, so changing a filter after the preview invalidates it.

Where does my API key go?

Into your MCP client’s own config file, on your machine. The server reads it from the environment and sends it to our API. It is never written anywhere else and never leaves your machine except as the Authorization header on a request to us.

Which mode should I use?

Safe, which is the default: your assistant can search, pull rows and look up domains, and anything expensive stops for your agreement. Use readonly for a shared or unattended assistant, and full only on your own machine when you want enrichment as well.

Does this need a special plan?

It needs an API key, which every paid plan has. The API is not available on the Free Trial, so the MCP server is not either.

Does it change anything about my account or the data?

No. It is a client, like the CLI or your own script. It reads what your key is allowed to read and spends only what you confirm. Nothing about the API changes to support it.

The API it wraps →Prefer a terminal? The CLI →See the data first →

Point your assistant at nine million stores

An API key comes with every paid plan, and the free trial shows you the data before you decide.

Start free See pricing