Sales intelligence MCP servers: overview, examples, and use cases

Six months ago, connecting a sales tool to an AI assistant meant writing custom API wrappers, parsing responses, and maintaining authentication flows for every provider. 

The Model Context Protocol (MCP) changed that equation. 

For developers building AI-powered sales workflows, MCP now removes the integration plumbing that used to sit between your model and your data. 

Instead of writing code to translate between Claude and ZoomInfo's API, you point Claude at ZoomInfo's MCP server (or Merge Agent Handler’s ZoomInfo server) and it handles the tool calls natively. 

To help you use sales intelligence MCP servers effectively, we’ll break down the ones available today and how development teams are combining them in production workflows.

How sales intelligence MCP servers work

Sales intelligence MCP servers expose tools to AI models through a standard protocol. 

Each server publishes a list of capabilities (e.g., search contacts, enrich companies, read CRM records) along with their input schemas. 

When a user asks Claude to "find Series B fintech companies in New York with 50-200 employees," Claude maps that request to the right tool call on the connected MCP server, sends the structured query, and returns the results.

Sales intelligence servers fall into three categories:

Data and enrichment servers 

These servers connect to B2B databases. They expose search endpoints (e.g., find companies) and enrichment endpoints (e.g., given a domain or profile URL, return structured firmographic and contact data). 

These are the highest-value MCP servers for sales teams because they replace the manual process of searching a database UI, exporting results, and pasting them into another tool.

CRM servers

CRM MCP servers give AI models read (and sometimes write) access to pipeline data. They let Claude pull deal history, contact records, and account associations without the user opening their CRM.

Engagement servers 

Engagement servers let you connect to outreach and sequencing platforms (e.g., Apollo). They handle the action layer: enrolling contacts in sequences, searching email history,  pulling call transcripts for meeting prep, etc.

Best sales intelligence MCP servers 

Here are some of the best MCP servers available across each category.

Data and enrichment

Crustdata

Crustdata's MCP server connects Claude to a database of 700M+ companies and 1B+ people profiles. 

The search layer is the primary differentiator: the company search endpoint supports 95+ filters with nested boolean logic, and the people search endpoint supports 60+ filters. 

Claude translates natural language ICP descriptions into filter combinations, which means non-technical users can build precise prospect lists without learning a query language.

Beyond search and enrichment, the server exposes social post retrieval (pull a prospect's recent posts for outreach personalization) and a web search endpoint. And the Watcher product pushes webhook notifications on events like job changes, funding rounds, and hiring spikes, though Watcher configuration happens through the REST API rather than MCP.

• Key tools exposed: Company search, people search, company enrichment, people enrichment, social post retrieval, web search
• Works with: Claude Desktop, Claude Code
• 3rd-party-hosted server: Merge Agent Handler

ZoomInfo

ZoomInfo's MCP server exposes six tools: find accounts, enrich accounts, research accounts, find contacts, enrich contacts, and research contacts. It connects to ZoomInfo's B2B database for company search by industry, revenue, and technographics, along with AI-ranked contact recommendations.

The setup is straightforward if you already have ZoomInfo API credentials. The constraint is access: ZoomInfo requires an existing subscription, and those start at enterprise pricing tiers. 

• Key tools exposed: Account search, account enrichment, contact search, contact enrichment, account research, contact research
• Works with: Claude Desktop, Claude Code
• 3rd-party-hosted server: Merge Agent Handler (supports significantly more tools out of the box!)

Apollo

Apollo's MCP server launched in February 2026. It lets Claude search for people and companies, enrich records, create or update contacts, and enroll prospects into outreach sequences. All actions sync back to Apollo as the system of record, and the server is available on all paid Apollo plans at no extra cost.

The server covers both data and engagement, which makes it versatile on paper. In practice, teams building serious prospecting workflows have reported mixed results with Apollo's search quality through MCP, and several have moved to dedicated data providers for the search and enrichment layer while keeping Apollo for sequence execution.

• Key tools exposed: Contact search, company search, enrichment, contact creation, sequence enrollment
• Works with: Claude (via Connectors)
• 3rd-party-hosted server: Merge Agent Handler

Amplemarket

Amplemarket's MCP server launched in March 2026. It exposes prospect search, contact enrichment, activity history, and list management. Authentication is browser-based (no API keys required), and it works with both Claude and ChatGPT.

• Key tools exposed: Prospect search, contact enrichment, activity history, list management
• Works with: Claude Desktop, ChatGPT

CRM

HubSpot

HubSpot's MCP server provides read-only access to CRM objects: contacts, companies, deals, tickets, invoices, products, quotes, and their associations. Setup uses OAuth through a HubSpot developer app or the HubSpot CLI.

The read-only limitation is significant for production workflows. You can pull pipeline data and account context, but you can’t update records or create new ones through MCP. Write capabilities are planned but not yet available.

• Key tools exposed: Read contacts, companies, deals, tickets, invoices, quotes, associations
• Works with: Claude Desktop, Claude Code
• 3rd-party-hosted server: Merge Agent Handler (it comes with tools that support a wide range of writes actions!)

Salesforce

Salesforce offers MCP support through the Salesforce DX MCP Server (Developer Preview). It lets Claude query data using SOQL, deploy metadata, and run tests. Agentforce 3 adds a governance layer for securely exposing internal APIs through MCP with Einstein Trust Layer controls.

This server is oriented more toward RevOps and admin workflows than frontline selling. Setup requires the Salesforce CLI and admin permissions; expect 15-30 minutes for initial configuration.

• Key tools exposed: SOQL queries, metadata deployment, test execution, Agentforce integration
• Works with: Claude Code, Claude Desktop (via community servers)
• 3rd-party-hosted server: Merge Agent Handler

Engagement

Outreach

Outreach became the first revenue orchestration platform to connect to Claude via MCP in February 2026. The server exposes tools for call transcript retrieval, email search, sequence lookup, prospect and account search, and deal queries.

Like HubSpot, it only supports read-only. You can pull pipeline intelligence and meeting context, but you can’t create sequences or enroll contacts. Access requires an admin to enable the MCP Server toggle in Outreach's organization settings, and each user needs an Amplify-enabled license.

• Key tools exposed: Call transcripts, email search, sequence lookup, prospect search, deal Q&A
• Works with: Claude Desktop, Claude Code

Web Research

Brave Search

Brave Search MCP gives Claude web search access without requiring an API key on the free tier. For sales workflows, it fills the qualitative gap that structured databases miss: recent news, product launches, and community discussions about a target company.

• Key tools exposed: Web search, news search
• Works with: Claude Desktop, Claude Code

Related: Examples of MCP servers

Sales intelligence MCP server use cases

Individual MCP servers are useful, but the real value shows up when you combine two or three into a workflow.

Here’s what some of these workflows can look like.

ICP-based lead discovery

Say a developer configures Claude with a data MCP server and a CRM server. 

A sales rep can describe their ideal customer profile: "Series B SaaS companies in North America, 100-500 employees, that have posted engineering job openings in the last 30 days." 

Claude then translates that into structured filters on the data server, returns matching companies, checks the CRM server to flag any that already exist as accounts, and outputs a net-new list with enriched firmographic data.

The entire flow happens in a single conversation. No CSV exports, no manual deduplication, no switching between tabs.

Meeting prep briefings

Imagine you’ve connected a CRM server, an engagement server, and a web search server. 

Before a call, a rep can then ask Claude: "Brief me on Acme Corp." 

Claude pulls the deal history and recent activity from the CRM, searches for recent email threads and call transcripts from the engagement platform, and runs a web search for any news from the past week. The output is a single briefing document built from three data sources in under a minute.

Signal-triggered outreach pipelines

Say a developer sets up a data provider's webhook or watcher system to push events (e.g., job changes, funding rounds, hiring spikes) into a queue. 

When an event fires, a Claude Code script picks it up, enriches the company and relevant contacts through MCP, checks the CRM for existing relationships, and either drafts personalized outreach or enrolls the contact in a sequence through an engagement MCP server.

The MCP layer handles the exploratory and prototyping phase of building this pipeline. For production execution at scale (hundreds of events per day), most teams move the enrichment and sequencing steps to direct API calls for better rate limit control and error handling. MCP is how you design the workflow; APIs are how you run it in production.

Best practices for using sales intelligence MCP servers

Before investing in sales intelligence MCP servers, here are some best practices worth following:

• Start with two servers. Scale later. Every MCP server adds tool descriptions to the model's context window. The more servers you connect, the more that window gets consumed by tool definitions rather than your actual conversation. Two servers (e.g., one data, one CRM) is the right starting point

• Separate data from action. The pattern that works best across teams we've spoken with: use a dedicated data provider for search and enrichment, and a separate server for CRM or outreach actions. Trying to use one server for everything usually means compromising on data quality or action capabilities

• Test before building. MCP's conversational interface makes it fast to evaluate data quality. Connect a server, run ten searches that match your real ICP, and check whether the results are accurate and fresh. Do this before investing time in building workflows on top of the data

• Plan the MCP-to-API transition. MCP is ideal for ad-hoc work and workflow prototyping. For production pipelines that process hundreds of records per day, direct API calls give you rate limit control, retry logic, and predictable costs. Design your architecture to support both: MCP for the exploratory layer, REST APIs for the execution layer
  

• Explore managed MCP connectors from integration platforms. For example, Merge Agent Handler’s sales intelligence MCP servers often support tools that the official servers don’t. In addition, Agent Handler comes with fully-searchable logs, customizable rules and alerts, and other features to keep your data secure and your workflows running smoothly

Final thoughts

The sales intelligence MCP ecosystem is still early. 

Most CRM servers are read-only. Some data servers return better results than others. Engagement servers vary in what actions they actually expose versus what they only let you read. 

Evaluating each server against your specific workflow matters more than picking the one with the longest feature list. And knowing when to use a 3rd-party-hosted MCP provider is equally important.

If you want to test a sales intelligence MCP server, Crustdata's API documentation covers setup for both Claude Desktop and Claude Code. Start with one data server, run ten queries against your real ICP, and evaluate the results before building workflows on top.