An MCP server is a small service that lets an AI assistant, such as Claude or another MCP-compatible tool, use an outside application through a standard set of actions. MCP stands for Model Context Protocol: instead of copying data into a chat window, the assistant calls the app's tools directly and works with real results.
Why this matters for outreach
Prospecting involves a lot of small, repetitive jobs. You search for businesses, check which ones have a website and an email, sort them, write a first line, and load them into a campaign. Each step is simple, but together they eat an afternoon.
An AI assistant can help with the thinking parts: deciding which leads fit, drafting openers, spotting gaps. Without MCP, though, it only sees what you paste in. You end up exporting a CSV, pasting rows into a chat, copying the answer back out, and hoping nothing got mangled.
With an MCP server, the assistant can ask the tool for the data itself. You describe what you want in plain language, and it runs the searches or reads the lists on your behalf. The data stays in the app where your campaigns live.
How it differs from a REST API
Both give software access to an app, but they serve different users.
| REST API | MCP server | |
|---|---|---|
| Who uses it | A developer writing code | An AI assistant acting on your request |
| How you ask | HTTP requests with parameters | Plain-language instructions in the assistant |
| Best for | Fixed, repeatable automations | Ad hoc research and one-off jobs |
If you want a scheduled job that runs the same way every Monday, an API script is usually the steadier choice. See REST API for leads for that route. If you want to ask questions and act on the answers as you go, MCP fits better.
A concrete example
Say you run a small web design agency and want to pitch dentists in three nearby cities.
GetClient offers an MCP server alongside its REST API. With it connected to your AI assistant, you could type something like: "Find dentists in Leeds, Bradford and York on Google Maps. Show me the ones with no website or an outdated-looking one, and list the owner's name where it's available."
The assistant calls the search, reads back business name, category, address, phone, website, any public emails and owner names with a confidence score. It can then sort them, flag the ones with no email, and suggest which to contact first. From there you might ask it to draft a short opener for each, using merge tags written exactly as {{params.firstname}} and {{params.website}}, so they fill in correctly when the campaign sends.
You still review the list and the copy. The assistant does the fetching and the first pass; you make the calls. For more on targeting this kind of client, see getting clients for a web design agency.
Common mistakes
Letting it run unchecked. An assistant can misread a request. "Dentists" might pull in dental labs or suppliers. Always look at the results before anything goes into a campaign.
Forgetting it spends credits. Searches and enrichment run through your account the same way they would if you clicked the buttons. Ask for a small batch first, check it, then scale up. Lead generation credits explains how this tends to work.
Trusting invented details. Assistants sometimes fill gaps with plausible guesses. If a field is empty in the data, it should stay empty in your outreach. Do not let a guessed owner name reach a real inbox.
Skipping verification. Run email verification before sending, whatever pulled the list together. Bounces hurt your sending accounts.
Writing merge tags loosely. If the assistant writes {{firstname}} or [Name], the field will not fill in. Correct it to the exact form before saving the step.
Using it for things that should be fixed. If you want the same list every week, a scripted job is more predictable than asking an assistant each time. See automating weekly lead lists.
What to do next
- Check that your AI assistant supports MCP. Not every chat tool does, and setup steps vary.
- Connect one MCP server and test it with a read-only question, such as listing a saved folder of leads.
- Try one small search in a single city, review every row, and note what the assistant got wrong.
- Write clear, specific prompts: category, location, what fields you need, and what to exclude.
- Keep a person in the loop for anything that sends email. Let the assistant prepare; you approve.
If you want to try this with GetClient, new accounts start with 100 free credits and no card. Details are on the plans and credits page.