What the MCP will let an agent do
Five things, in the words an agent already uses.
The design is deliberately small. An agent should not have to learn a filter schema to ask for dentists in Rotterdam, and it should never spend credits to find out whether a search was worth running. These are the capabilities the MCP is being built around, and they are in early access rather than installable today.
- Search businesses in plain language: a niche and a place, the way you would say it out loud.
- Search people by role, industry and location, for the decision maker rather than the storefront.
- Read results: name, website, phone, address, category, rating and review count from the Google Maps listing.
- Get the verified contact: the email taken off the business's own website while the search runs, and the work email of the person where one is found.
- Preview the count for free, so the agent can tighten a filter, widen a city or stop before anything is charged.
That last one matters more than it sounds. An agent in a loop is the easiest way ever invented to spend money on a bad query. A free count turns the first step of the loop into a question rather than a purchase.
Which clients it is for
Any MCP client, with the obvious ones named.
MCP is a standard, so the target is any client that speaks it. In practice the ones people ask about are Claude Code, Claude Desktop, Cursor, n8n and Make, and those are the ones early access is aimed at first. A Claude Code skill is being built alongside it so an agent picks up the workflow rather than just the tool list.
This is not a new piece of infrastructure invented for LeadFindr. DM Champ already runs a hosted MCP server for its own API, and LeadFindr joins it. That is the one true supporting fact here, and no install command, endpoint or tool list is published yet because none of it is live.
How a search will run
Start, poll, fetch, because the checking happens live.
A database MCP answers in one call, because it is reading rows that already exist. LeadFindr cannot do that honestly: it opens each business's own website while your search is running, which takes longer than a lookup and is the entire reason the contact details are current.
So the shape of an agent search is start, poll, fetch. The agent starts a search with a cap it chose, polls while results stream in, and fetches the finished list. An agent handles that pattern far better than a human staring at a spinner, which is part of why this product suits agents at all.
What it costs an agent
The same price a person pays, with a free look first.
A lead with contact details costs 0.25 credits, about $0.025 at the $0.10 top up price of a credit. A lead with no email, no phone number and no social profile costs nothing. The count preview costs nothing. Every search carries a cap the agent sets before it runs, so a runaway loop hits a ceiling rather than a balance.
There are no per seat fees and no subscription, which is the part that matters when the thing doing the searching is software rather than a person. You are not buying a seat for a bot. New accounts start with 100 credits, which is 400 leads.
What is live today, and what is not
The honest split, so nobody builds against something that does not exist.
- Live today: the app. Search local businesses and decision makers, live website checks, verified emails, saved searches, starred lists, CSV export, import into DM Champ as contacts.
- Early access: the MCP tools, agent API keys and the Claude Code skill. Not installable, no published endpoint, no tool list.
- Not built: a public REST API. If your pipeline needs one today, this is not the tool yet.
Request early access and you will be told where it stands. We would rather be the product you waited two weeks for than the one whose docs described software that was not there.
Questions People Actually Ask.
Can my AI agent use LeadFindr today?
Through the app, yes, a person runs the search. Agent access over MCP is in early access: request it and you will be told where it stands. There is no install command or endpoint published yet.
Which MCP clients will be supported?
Any client that speaks MCP. Early access is aimed first at Claude Code, Claude Desktop, Cursor, n8n and Make. DM Champ already runs a hosted MCP server for its own API and LeadFindr joins it.
Is there a REST API?
No. There is no public REST API today, and no date is being promised for one. MCP is the route being built.
What does a search cost my agent?
0.25 credits for each lead that comes back with contact details, about $0.025. Leads without contact details are free, the count preview is free, and every search carries a cap the agent sets before it runs.
Does it work in Claude Code?
That is the first client early access targets, together with a Claude Code skill so an agent picks up the whole workflow. It is not installable yet.
Why does a search take longer than a database query?
Because each business website is opened while the search runs instead of read from a stored crawl. That is what keeps a closed business or a dead site out of the results.