If you’ve built anything on top of an LLM in the last six months, you’ve hit the same wall: the model is smart, but it can’t see your data. Every sales tool in your stack is behind a different API, with different auth, different rate limits, different error handling. The agent can’t use any of it without a custom integration.
I’m Nicolas, head of partnerships at Leadpipe. MCP is the open standard fixing that, and sales data is one of the cleanest use cases. This is a field guide to MCP servers for sales data: what they are, why they matter, how to evaluate one, and what the Leadpipe MCP server exposes.
What MCP is, in one paragraph
MCP stands for Model Context Protocol. It’s an open standard, created by Anthropic, for connecting AI agents to external tools and data. Instead of writing custom API integration code for every agent and every tool, a tool vendor publishes an MCP server. Any MCP-compatible agent, Claude Desktop, Cursor, Codex, or a growing list of frameworks, can plug that server in and immediately get structured access to every tool it exposes.
Think of it as a USB port for AI. Before USB, every peripheral had its own connector. After USB, you plug anything into anything. MCP is doing that for agents and data.
Why sales data is the use case
Sales data has three properties that make it an ideal MCP use case.
| Property | Why it matters for MCP |
|---|---|
| It’s structured | Agents need structured tools with clear inputs and outputs, and sales data fits that shape naturally |
| It’s high-leverage per call | One lookup can change the outcome of a rep’s day, so agents can justify the tool call |
| It’s fragmented across vendors | Every sales stack has six to ten tools, and agents need a unified way to reach them |
A generic web-search MCP is useful. A sales-data MCP is transformative because it turns the agent from a conversational layer into an operator. The agent can look up visitors, build audiences, pull intent topics, check pixel status, and act on the results, all in one conversation.
For the broader argument on why every AI agent needs identity data, see why every AI agent needs an identity API and the data layer AI sales agents are missing.
What a good sales-data MCP exposes
Not every vendor that ships an MCP server ships a useful one. A minimal MCP that just wraps a search endpoint isn’t very different from a web-search plugin. A useful sales-data MCP covers four categories of tools.
| Category | What it enables | Example tool |
|---|---|---|
| Discovery | Find topics, accounts, or people matching a description | search_topics, analyze_website_topics |
| Audience building | Construct and save audience definitions with filters | preview_audience, create_audience |
| Retrieval | Pull actual person or visitor records | query_visitor_data, get_audience_results |
| Operations | Manage the underlying infrastructure (pixels, credits) | list_pixels, get_account_status |
If the MCP skips any of these, the agent has to fall back to the dashboard for whatever’s missing, which defeats the point.
The Leadpipe MCP, at a glance
Our MCP server, @leadpipe/mcp, ships 27 tools across six categories. One line installs it:
npx -y @leadpipe/mcpThe tool inventory:
Topic discovery (7 tools)
Browse the 20,810-topic taxonomy, search by keyword, get trend data, compare topics, find movers, and classify arbitrary URLs into topics.
Audience builder (8 tools)
Get filter options, preview audience size before committing credits, run queries, save and update audience definitions, activate or pause them.
Audience results (5 tools)
Check materialization status, fetch unmasked person records, list daily run history, inspect field fill rates, and generate CSV exports.
Visitor data (1 tool)
Look up identified visitors by email, domain, or timeframe.
Pixel management (3 tools)
List, create, and update tracking pixels, including excluded-path configuration.
Account (1 tool)
Get credits remaining, pixel count, audience slots, and an overall health check.
Plus three built-in prompts that walk agents through the common workflows end to end: topic discovery, audience operations, and visitor investigation. The full tool table lives in the MCP server launch post.
Setup in under five minutes
The install is the same three-step pattern across every MCP client. You add a small JSON config pointing at npx -y @leadpipe/mcp with your API key as an environment variable. You restart the client. All 27 tools appear.
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"leadpipe": {
"command": "npx",
"args": ["-y", "@leadpipe/mcp"],
"env": {
"LEADPIPE_API_KEY": "sk_..."
}
}
}
}Cursor (~/.cursor/mcp.json): same shape as above.
Codex (OpenAI CLI):
codex mcp add leadpipe --env LEADPIPE_API_KEY=sk_... -- npx -y @leadpipe/mcpAny MCP-compatible client: the server runs over stdio. Point any MCP client at npx -y @leadpipe/mcp with LEADPIPE_API_KEY in the env.
No OAuth flow, no dashboard configuration, no separate install. If your API key works, the MCP server works.
What the agent can actually do
The setup is mundane. What the agent does afterward is where the leverage is.
Topic discovery in natural language
Find strong B2B intent topics for procurement automation software.
Use Leadpipe topic discovery to search, compare recent trend lines,
identify fast-growing movers, and recommend the best 5 topics for
an audience preview. Explain why each topic made the cut.The agent walks through search_topics, compare_topics, get_topic_movers, and preview_audience on its own. You read the summary.
Audience building without touching the dashboard
Preview an audience for B2B topics related to cloud security
posture management. Require business email and LinkedIn, company
size mid-market or enterprise, minimum score 70. Return estimated
size, a short interpretation of the sample, and any warning if the
audience looks too narrow.The agent calls get_audience_filters to see what’s available, then preview_audience with the constraints, then summarizes.
Visitor investigation from a conversation
Investigate visitor data for example.com over the last 30 days.
Check account health, query visitor activity, inspect configured
pixels, and tell me whether tracking looks healthy.The agent hits get_account_status, query_visitor_data, and list_pixels in sequence, then gives you one paragraph of signal.
None of this requires code. None of it requires the agent to know Leadpipe’s API schema. MCP handles the discovery and invocation; the agent just picks the right tool for the prompt.
For the full prompt library, see the MCP launch post and the prompt examples on GitHub.
When to use MCP vs raw API
A fair question. We also ship 23 REST endpoints and a TypeScript SDK (npm install @leadpipe/client). When do you use which?
| Raw API / SDK | MCP Server | |
|---|---|---|
| Who uses it | Your code | Your AI agent |
| Integration effort | Write code | Add 5 lines of config |
| Authentication | Handle in code | Configured once in env |
| Error handling | Your responsibility | Agent handles retries |
| New tools | Update your code | Update the npm package |
| Works with | Any language | Claude, Cursor, Codex, any MCP client |
The rule we use internally: if you’re building a product on top of Leadpipe data, use the REST API or the SDK. If you want an AI agent, a rep, or a founder to have direct conversational access to Leadpipe data, use the MCP server.
For the broader comparison on how identity-data APIs fit into AI workflows, see identity resolution API for AI agents and self-serve identity resolution API.
How to evaluate any sales-data MCP
If you’re looking at other MCP servers in the sales data category, here’s a quick checklist. A production-ready MCP should answer yes to all of these.
- Is it npx-installable? Local install with no server hosting burden on you.
- Does it cover all four tool categories? Discovery, audience/query building, retrieval, and operations. Missing any one forces you back to the dashboard.
- Does it include built-in prompts? The best MCP servers ship prompt templates that walk the agent through common workflows without the user having to write them.
- Is it open source? MIT or similar license means you can read the code and understand what your agent is actually calling.
- Does it pass API errors through cleanly? If the agent gets garbage on a rate limit, it’ll give you garbage back.
- Is the API key scope clear? Environment variable only, never prompted mid-conversation.
The Leadpipe MCP passes all six. The npm package is MIT-licensed, the GitHub repo is public, and the full tool inventory is documented.
What this means for customers
Most sales tools have an API. Few of them have an MCP server. The ones that do are usable inside your agent today. The ones that don’t will be usable as soon as you write and maintain a custom integration, which is not a free lunch.
Leadpipe is in the first category. The 27-tool MCP server is shipped, open source, and free to install. You only pay for the underlying API usage, which follows the same plan pricing as the REST API: $147/mo Pro, $1,279/mo Agency white-label, or custom enterprise.
If you’re building anything with Claude, Cursor, Codex, or any MCP-compatible agent, and you need sales data in the mix, the MCP server is the fastest path from “agent that can chat” to “agent that can operate.”
Every plan ships with the same identity graph, 23 REST endpoints, webhooks, and a 27-tool MCP server. Start in 5 minutes
