Most explanations of MCP are written for developers. This page is for the sales or GTM person who already works in Claude Code, Cursor or Codex and wants to run prospecting from there: what the protocol is, what a sales server exposes, how a task runs, what stays in the platform, and what to check on credits, permissions and security.
What is an MCP server, in one paragraph?
An MCP server is a program that offers data and actions to AI applications through the Model Context Protocol, an open standard introduced by Anthropic in November 2024. The specification describes three roles. The host is the AI application you work in, such as Claude Code or Cursor. Inside it, a client keeps a connection to each server. The server offers resources, prompts and tools; when the model needs one, the client calls it and passes the result back. For a sales team, the server is the bridge between the AI tool you already use and the platform that holds the data.
What does an MCP server expose for a sales team?
The platform's core actions, each as a tool the model can call. A typical sales server exposes five:
| Tool | What it does | Example request |
|---|---|---|
| Search | Finds companies or people in the lead database by criteria | SaaS companies in Germany with 50 to 200 employees |
| Lists | Creates a list and fills it from search results | Save the matches as a new list |
| Enrich | Adds verified emails, phone numbers and company data | Find work emails for the heads of marketing |
| Sequence | Creates a sequence and adds contacts to it | Add the verified contacts to a two-step email sequence |
| Credit cost | Shows what an action will cost before it runs | What will enriching these contacts cost? |
Resources and prompts can add context, such as the team's ideal customer profile, but tools do most of the work.
The MCP specification defines hosts, clients and servers, the JSON-RPC 2.0 message format, and three server features: resources, prompts and tools. Its security section asks hosts to obtain explicit user consent before invoking tools. Governance changed in December 2025, when Anthropic donated the protocol to the Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI; by then there were more than 10,000 public MCP servers.
How does a prospecting task run from Claude Code or Cursor?
As a conversation in which the model calls tools one after another. A request such as 'find heads of marketing at SaaS companies in DACH with 50 to 200 employees, verify their emails and put them in a new sequence' runs roughly like this:
- The host reads the request and checks which tools the connected servers offer.
- Search returns matching companies and people.
- Credit cost reports what enrichment will cost, and the model asks you to confirm.
- Enrich adds and verifies email addresses.
- Sequence creates the sequence and adds the verified contacts.
- The host summarises what was found, what it cost and what is ready to send.
Each step is a tool call you can see in the conversation. The specification expects the host to obtain your consent before a tool runs, so spending credits or launching a sequence is a decision you approve, not a side effect.
What stays in the platform, and what moves to the IDE?
The data, the sending infrastructure and the rules stay in the platform. The orchestration moves to the IDE.
- Stays in the platform: the lead database, enrichment providers, mailboxes and sending schedules, the shared inbox, CRM sync and credit accounting.
- Moves to the IDE: the order of steps, the logic between them, and anything that combines the platform with other systems, such as your own product data or a classification script.
Leadspicker's knowledge base describes the point of this split: use exactly the slice you need, the data, the enrichment or the sending, inside a larger pipeline you build yourself, so that Leadspicker is one step in the workflow rather than the workflow. Teams without someone to own a pipeline in code lose nothing: the same actions remain available in the app.
What about credits, permissions and security?
Check three things before connecting any sales MCP server.
- Authentication: prefer OAuth with your own account over a shared API key pasted into a config file. OAuth ties every call to a person and can be revoked.
- Scope: the server should read and change only your own workspace, with the same permissions you have in the app.
- Cost: know whether calls spend the same credits as the app or a separate allowance, and ask the model to check the cost before large enrichment runs.
The specification is explicit that tools can run arbitrary actions and must be treated with caution. Hosts should obtain explicit user consent before invoking a tool, and descriptions of tool behaviour should be treated as untrusted unless they come from a trusted server. Connect only servers from vendors you already trust with your data.
Leadspicker's MCP server is remote and signs in with OAuth instead of an API key. It reads and changes only the user's own workspace, spends the same credits as the app, and is included on all plans. It connects to Codex, Cursor, Claude Code and other MCP-capable tools.
Frequently asked questions
What is an MCP server for sales teams?
A server that exposes a sales platform's actions, such as search, lists, enrichment and sequences, as tools an AI assistant can call through the Model Context Protocol. You work in the AI tool; the platform does the work in the background.
Do MCP calls cost extra credits?
It depends on the server. Leadspicker's MCP server spends the same credits as the app, so an enrichment costs the same whether you run it from Cursor or from the app. Ask the model to check the cost before large runs.
Is it safe to connect a sales MCP server to an AI coding tool?
It can be, with three checks: OAuth sign-in with your own account, access limited to your own workspace, and a host that asks for consent before each tool call. Connect only servers from vendors you already trust with your data.
- Anthropic, Introducing the Model Context Protocol (2024)
- Model Context Protocol, Specification (2026-07-28)
- Anthropic, Donating the Model Context Protocol and establishing the Agentic AI Foundation (2025)
- Leadspicker, MCP server page and pricing (October 2026)
- GTM playbook for 2026 by Leadspicker, chapters: Using Leadspicker from your AI tools (MCP)
