For a long time, “using AI at work” meant a slightly awkward routine. You’d describe a task to the chatbot, copy its answer, flip to another tab, and do the real work yourself. The Model Context Protocol, or MCP, has quietly changed that. You connect a tool to the assistant, and it calls that tool for you, right there in the conversation.
Many developers first ran into MCP inside code editors like Cursor. Now sales and marketing teams are finding it handy for one of their most mind-numbing jobs: building and cleaning prospect lists. Plug an email finder MCP server for Claude into the assistant, and “find me forty work emails” becomes a request instead of an afternoon.
So what is an MCP server, exactly?
It’s a small service that offers a set of tools to an AI app in a standard format. The app might be Claude.ai, Claude Desktop, Claude Code, Cursor, Windsurf, or something else. It discovers those tools, and the model decides when to use them based on what you ask. There’s no plugin store to wait on and no copy-pasting between apps. The tool runs, the result lands back in the chat, and the assistant can keep going: sorting it, summarising it, or turning it into a table.
Servers usually come in one of two flavours. Hosted servers get added as a remote connector with a URL and a sign-in. Local servers run on your own machine, often as an npm package you start with a single npx command, and they read an API key from a config file. There’s also an official MCP registry, launched in preview in September 2025, that lists published servers. It’s a useful first stop for checking what a server actually is before you connect it.
Why prospecting is such a good fit
Picture Sam, a made-up sales rep at a mid-sized software company. Her week is a chain of tiny lookups. Who runs sales at this company? What’s their work email? Is it safe to send to? None of those steps is hard on its own. But repeat them for forty names and the afternoon’s gone.
That’s exactly the kind of work an assistant can orchestrate once it has the right tools. It’s also work where Sam needs a real answer, not a guess.
That last bit matters more than it sounds. Ask a language model on its own for Jane Cooper’s email at Acme, and it’ll cheerfully offer jane.cooper@acme.com. It might be right. It might bounce and dent your sender reputation. An email finder connected over MCP swaps that guess for an actual lookup and, ideally, a verification step.
What it looks like in practice
Take Scrupp’s server as an example. In Claude.ai or Claude Desktop, Sam goes to Settings, then Connectors, then Add custom connector, and pastes in the server’s URL. In Cursor and other clients, it runs with npx -y scrupp-mcp-server an API key. After that, she just asks in plain English:
“Find the work email of Jane Cooper at acme.com.”
“Here are 40 names with company domains. Find their emails and give me a table.”
“Check these five addresses before I send.”
The assistant calls the find or verify tool, waits for bigger jobs to finish, and drops the results into the chat.
Now for the honest part: the hit rate. No email finder covers everyone, and the decent ones admit it. By Scrupp’s own measurement across all exports, a work email comes back for an accepted delivery confirms 36.3% of rows, and 10.0%. In other words, Sam’s list of forty might give her around fourteen emails, and only a handful will be fully confirmed. Lookups are only charged when an email is found, though, so the cost follows the results, not the length of the list.
Clean the list before the AI sees it
Lots of prospect lists start life in LinkedIn Sales Navigator, and Sales Navigator links have an annoying habit of breaking downstream. A lead link looks like linkedin.com/sales/lead/ACwAAA... and only opens for someone with a Sales Navigator seat. That’s a problem for your CRM, your enrichment tools, and the teammate who doesn’t have a licence.
The fix is simple once you know it. The chunk starting with ACw is the person’s member ID. Put it after linkedin.com/in/ and you get a regular profile link that anyone can open. A free Sales Navigator URL converter does this for a whole list at once, right in the browser. If your links already live in Google Sheets or Excel, you can do the same job with a formula.
It’s worth sorting this out before the assistant ever touches the list. Clean, consistent inputs- meaning a name, a company domain, and a working profile link- are what let an MCP tool give you solid results instead of half-finished ones.
A few habits worth keeping
Giving an assistant real tools is powerful, so a little caution goes a long way.
Check where a server comes from. Stick to ones listed in the official registry and published by a vendor you already know. Remember that connecting tools also gives an assistant a way to reach out across the network, and that’s why AI agent security is becoming its own discipline.
Know how billing works. Find out whether you pay per attempt or per result, and keep an eye on your credits. The better servers let the assistant check your balance for you.
Verify before you send. Even a verified address should pass a deliverability check before it enters a sequence.
And respect data protection rules. Collect only what your outreach genuinely needs, and make unsubscribing easy.
Where this is all heading
MCP is less about any single tool and more about where the work happens. Research, lookups, and clean-up all move into the same conversation where you plan the campaign, and the assistant stitches them together. It’s part of a broader shift towards AI marketing agents handling the legwork while people keep the final say. For prospecting, that means fewer tabs, fewer copy-paste mistakes, and lists built from real lookups instead of confident guesses. The teams getting the most from it pair the convenience with realistic expectations about hit rates, and they verify before they hit send.
Related: 11 Best Agentic AI Frameworks in 2026: A Complete Decision Guide
