Using the MCP server with Claude and other AI clients
Point an AI client at your account and ask for a list in plain language. What the tools are, what they cost, and the guardrails.
The MCP server exposes search, filtering and export as tools an AI client can call. Point Claude, ChatGPT, Gemini or any MCP-compatible client at your account and ask for "dentists in Berlin with a website but no email" in ordinary language.
Like the REST API it sits alongside, the MCP server starts on Growth. It is not on Starter; it is included in the free trial, which has everything unlocked. It runs on the same credits, the same permissions and the same per-second rate limit as your account's REST key.
Connecting
- Create an API key in Account → Security → Access keys.
- Add the FindAndClose MCP server in your client's connector settings.
- Paste the key when asked. The exact wording of these screens differs per client; the current per-client instructions are in Account → Security, next to the key you just made.
Treat the key as a credential, not a setting
An MCP client holding your key can spend your credits. Give each client its own key so you can revoke one without breaking the others — rotating and revoking keys.
What the agent can do
| Tool | Does | Costs credits |
|---|---|---|
| Look up a country, region or city | Resolves names to the ids a search needs | No |
| Look up business categories | Searches the 4,000+ categories | No |
| Search | Runs a query and returns the count plus masked rows | No |
| Create an export | Queues a job in your account | Yes |
| Check an export | Reports status and progress | No |
| Download an export | Returns the finished file | No |
Because searching is free, an agent can iterate — try a category, look at the count, try a broader one — without spending anything. Only the export step bills.
The guardrails
- It runs on your credits and your permissions. An agent cannot spend more than you could. It can reach every country and category, because every plan can — the ceiling it cannot pass is your credit balance.
- Search results are masked, to the agent as much as to you. It cannot read contact details out of a preview and paste them into a chat — why previews are masked.
- Exports are visible. Everything an agent creates appears in Account → Exports with its name and credit cost, so you can see what it did.
- Rate limits apply exactly as they do to the REST API.
Prompts that work well
- "How many independent bookshops are there in each of the five largest Norwegian cities?" — free, several searches.
- "Find dentists in Oslo with an email and at least 4 stars, tell me the count, and wait for me to confirm before exporting." — the confirmation step is worth asking for.
- "Export that as CSV with only the name, email and phone column groups."
Prompts that work badly are the ones that skip the count: "export every restaurant in France" will do exactly that, and the credits are real.
If a tool call fails
The errors are the REST errors — 401 unauthenticated for a bad key,
403 forbidden when the account is on Starter and has no MCP access, or when the
agent asked for an export format your plan does not include, 402 credits_exhausted
when you are out. Most clients surface the
message field verbatim. See
rate limits and error codes.
Updated on: