What is an MCP server?

An MCP server is a program that exposes tools, data or prompts to an AI assistant through the Model Context Protocol (MCP), an open standard. The assistant, called the MCP client (for example Claude, ChatGPT or Cursor), discovers what the server offers and calls its tools during a conversation, so it can query databases, read files or fetch live data without custom integration code.

Facts about other products checked on 2026-11-09 · sources listed below

	Local MCP server	Remote MCP server
Where it runs	On your own machine	On the provider's infrastructure
Transport	Standard input/output (stdio)	HTTP, reached by a URL
Setup	Install and run a package	Add a URL to your client
Sign-in	Usually local config or env vars	OAuth or an API key
Typical use	Local files, dev tools, private scripts	Hosted data and SaaS products
Example	A filesystem or database server	`https://mcp.leadsets.ai`

How an MCP server works

The Model Context Protocol defines a common way for AI applications to talk to outside systems. It has three roles:

Host: the app you use, such as Claude, ChatGPT or Cursor.
Client: the part of the host that holds a connection to one server.
Server: the program that offers capabilities, such as a database, a SaaS product or a data provider.

When a client connects, it asks the server what it can do. The server answers with a list of tools, each with a name, a plain-language description and a JSON schema for its inputs. The model reads those descriptions, decides which tool fits the user's request, and the client sends the call. The server runs it and returns structured results the model can reason over. Messages use JSON-RPC 2.0, so any client that speaks MCP can use any server that speaks MCP.

The practical effect is that one integration works across many assistants. A provider builds one MCP server instead of a separate plugin for every AI product, and users connect it wherever they already work.

What an MCP server can expose

MCP servers offer three kinds of capability:

Tools: actions the model can call, such as search_people or create_ticket. Tools are the most common capability and the one most people mean when they talk about MCP.
Resources: read-only data the client can load into context, such as a file, a schema or a document.
Prompts: reusable templates the server suggests for common tasks.

A good server keeps tools narrow and well described. The model chooses tools from their descriptions, so a tool that does one clear thing and returns predictable fields is easier for an agent to use correctly than a tool that does many things.

Local and remote MCP servers

The table above summarises the two main ways a server runs. A local server is a process on your computer that talks to the client over standard input and output. It suits private files and developer tools, but each user has to install and update it. A remote server runs on the provider's side and is reached over HTTP with a URL. There is nothing to install, sign-in usually happens through OAuth or an API key, and updates reach everyone at once.

Most hosted products, including data providers, ship remote servers. Several B2B data companies now publish one. Apollo says its remote MCP server (mcp.apollo.io) is available on any plan including Free. Lusha, Wiza (mcp.wiza.co) and ZoomInfo (mcp.zoominfo.com) each document an official MCP server, and Clay offers an official Clay MCP. In each case the vendor says MCP use draws on the same plan, credits or licence as its other access; ZoomInfo, for example, says its MCP server requires a ZoomInfo licence. Leadsets runs as a remote MCP server at https://mcp.leadsets.ai with eight tools. For a side-by-side view of these options, see B2B data APIs for AI agents.

An example: an MCP server for B2B data

Maya Chen runs revenue operations at Northwind Analytics (northwind.co). She wants 20 heads of RevOps at US fintechs with 50 to 200 employees, with work emails she can send to.

Without MCP, she would export a list from a data tool, clean it in a spreadsheet and paste it back into her assistant. With a B2B data MCP server connected to Claude, she types the request in plain language. Claude sees the server's tools and plans the run itself:

get_usage to check how much allowance is left.
search_people with filters for function, seniority, company size, industry and country.
enrich_person for the current role and a work email.
verify_email to label each address before anything is sent.

Claude returns a table and explains what it could not confirm, for example addresses on catch-all domains. Maya never writes code, and the same server would work if she switched to Cursor or ChatGPT. The tool names here are the ones Leadsets uses; other providers name and scope their tools differently, which is worth checking before you connect one. More on this workflow is on the Leadsets for AI agents page.

MCP server or API: what is the difference

An API is built for developers: you read the docs, write code, handle authentication and parse responses. An MCP server is built for AI clients: it describes its own tools so a model can discover and call them without custom code. Many providers offer both, often with the MCP server sitting on top of the same API and the same limits.

Use the MCP server when a person is working through an assistant and wants results in the conversation. Use the API when you are building a scheduled pipeline, a product feature or a workflow in a tool like n8n, where you want fixed logic rather than a model deciding each step. MCP servers also pair well with skills, which are written playbooks that teach an agent which tools to call and in what order; see what are Claude skills.

How to connect one, and what to check first

In most clients, adding a remote server takes a minute. In Claude you add a custom connector with a name and the server URL, then sign in. In Claude Code it is one command, for example claude mcp add --transport http leadsets https://mcp.leadsets.ai. In Cursor you add the URL and an authorization header to its MCP settings. Step-by-step instructions for each client are in the MCP connection guide.

Before connecting any server, check:

Who runs it. A server can read what you send it and act on your behalf, so connect only servers from providers you trust.
What its tools can do. Read-only tools carry less risk than tools that write, send or delete.
How usage is billed. MCP calls usually draw on the same credits, records or licence as the provider's API, and an agent can make many calls quickly.
What comes back. Structured fields with sources, dates and confidence make it easier for the model, and for you, to judge what to trust.

Questions and answers

What does MCP stand for?

MCP stands for Model Context Protocol. It is an open standard that defines how AI applications connect to outside tools and data sources through a common client and server interface.

Is an MCP server the same as an API?

No. An API is designed for developers who write code against it, while an MCP server describes its own tools so an AI client can discover and call them directly. Many providers offer both, and the MCP server often uses the same underlying API and limits.

Do I need to install anything to use an MCP server?

Only for local servers, which run as a process on your computer. Remote MCP servers are reached by a URL, so you add the address to your client and sign in with OAuth or an API key.

Which AI apps can use MCP servers?

Claude, Claude Code, ChatGPT and Cursor all support connecting MCP servers, along with many other agents and IDEs. Any application that implements the MCP client side can use any compliant MCP server.

Are MCP servers safe to connect?

An MCP server can see the requests you send it and may take actions on your behalf, so safety depends on who runs it and what its tools can do. Connect servers only from providers you trust, prefer read-only tools where possible, and check how usage is billed before letting an agent run.

What is the difference between an MCP server and a Claude skill?

An MCP server gives an agent tools it can call, such as a search or an enrichment. A skill is a written playbook, usually a SKILL.md file, that teaches the agent how to combine those tools step by step for a task like building a prospect list.

Sources · checked 2026-11-09

Prices and features change. If something here is out of date, email support@leadsets.ai and we'll fix it.

Go deeper