HubSpot MCP Server: How to Connect AI to Your CRM

In this guide

1. What a HubSpot MCP server does
2. Why this matters for revenue teams
3. How to set it up (step by step)
4. Available tools and capabilities
5. Real use cases by role
6. HubSpot MCP server vs. HubSpot API
7. Security and permissions
8. FAQ

What a HubSpot MCP server does

Instead of clicking through HubSpot's interface to find data or update records, you talk to an AI assistant. The MCP server translates your request into HubSpot API calls and returns the results.

The interaction feels like talking to a teammate who has instant access to your entire CRM:

• "Show me all deals closing this month over $50K" — queries your deal pipeline and returns a filtered list
• "Create a contact for Jane Smith at Acme Corp with this email" — creates the record with the properties you specify
• "What's our pipeline value by stage?" — pulls deal data and summarizes by stage
• "Find all contacts who opened our last email campaign but didn't click" — queries engagement data and returns the segment
• "Update the deal stage for Acme Corp to Contract Sent" — updates the deal record directly
• "Log a call with John at Widget Co — discussed pricing, follow up Thursday" — creates an engagement record with notes

Why this matters for revenue teams

HubSpot has over 228,000 customers across 135 countries. Most of those teams share the same problem: CRM data quality. Reps don't update records because it's tedious. Managers can't trust reports because the data is stale. Ops teams spend hours on manual data entry and cleanup.

An MCP server changes the interaction model. Updating your CRM becomes a conversation, not a chore. But the bigger impact is on bulk operations:

• Re-scoring an entire pipeline segment based on new criteria
• Enriching 500 contacts with missing company data
• Generating a weekly pipeline report with stage-by-stage breakdown
• Flagging every deal that hasn't moved stages in two weeks
• Building a list of contacts who match an ideal customer profile

Work that takes an ops person hours takes the AI minutes. And unlike a dashboard, you can ask follow-up questions: "Now filter that to just enterprise deals" or "Who's the owner on each of those stale deals?"

How to set it up

Step 1: Create a HubSpot Private App

Go to Settings → Integrations → Private Apps in your HubSpot account. Create a new app and select the scopes you need. For a read-only start, select:

• crm.objects.contacts.read
• crm.objects.companies.read
• crm.objects.deals.read

Copy the access token. You'll need it in the next step. See HubSpot's Private Apps documentation for details on scopes.

Step 2: Add the MCP server to your AI client

HubSpot ships an official MCP server as an npm package. Add it to your MCP client configuration (Claude Desktop, Cursor, VS Code, etc.):

{
  "mcpServers": {
    "hubspot": {
      "command": "npx",
      "args": ["-y", "@hubspot/mcp-server"],
      "env": {
        "HUBSPOT_ACCESS_TOKEN": "pat-na1-your-token-here"
      }
    }
  }
}

For Claude Desktop, this goes in ~/Library/Application Support/Claude/claude_desktop_config.json on Mac or %APPDATA%\Claude\claude_desktop_config.json on Windows.

Step 3: Test it

Restart your AI client. Ask something simple like "How many contacts do I have in HubSpot?" If the server is connected, it'll query the CRM and return the count.

Alternative: Community-built servers

Several open-source HubSpot MCP servers exist with different feature sets. Check the MCP servers repository on GitHub for the latest options. Community servers sometimes cover features the official one doesn't yet support, like custom object types or advanced workflow triggers.

Available tools and capabilities

A HubSpot MCP server typically exposes tools organized by CRM object:

• Contacts — create, read, update, search, list, bulk operations
• Companies — create, read, update, search, associate with contacts
• Deals — create, read, update, move between pipeline stages, associate with contacts and companies
• Engagement — log emails, calls, meetings, notes against records
• Lists — query static and active lists, add or remove members
• Properties — read custom property definitions, create new properties
• Search — full-text and filtered search across all object types

The exact tools available depend on which server you use and what scopes your access token includes. The official server covers the core CRM objects. Community servers may add marketing, automation, or reporting tools.

Real use cases by role

Sales reps

Log calls and meetings by describing what happened instead of filling out forms. Get instant pipeline summaries before manager 1:1s. Find contacts that match specific criteria without building HubSpot filters. Draft follow-up emails with deal context pulled from the CRM automatically.

Sales managers

Run weekly pipeline reviews by asking for stage-by-stage breakdowns. Identify stale deals across the team without building reports. Compare rep performance by pulling activity metrics. Get alerts on deals that need attention.

Revenue operations

Bulk-update properties across hundreds of records. Audit data quality — find contacts missing emails, companies without industry tags, deals without close dates. Build and test list segments using natural language criteria instead of HubSpot's filter builder.

Marketing

Pull campaign engagement data and segment respondents. Find contacts who match ideal customer profiles for targeted campaigns. Analyze which lead sources produce the highest-value deals by querying deal + contact data together.

HubSpot MCP server vs. HubSpot API

The HubSpot API already exists. So why use an MCP server instead?

• The API requires code. Every query needs an HTTP request with the right endpoint, headers, and parameters. An MCP server lets you use natural language.
• The API is point-to-point. You build an integration between one app and HubSpot. An MCP server works with any MCP-compatible AI client — Claude, ChatGPT, Cursor, VS Code, and more.
• The API doesn't reason. It returns raw JSON. An MCP server + AI model returns interpreted results: summaries, analysis, recommendations.
• The MCP server uses the API under the hood. It's not a replacement — it's a layer on top that makes the API accessible to non-developers and AI agents.

Security and permissions

Your MCP server has exactly the same access as your Private App token. Scope it tightly:

• Start read-only. Grant only .read scopes until you're comfortable with AI-initiated writes.
• Use a dedicated Private App. Don't reuse tokens from other integrations — if you revoke MCP access later, you don't break anything else.
• Audit the server source code. The official HubSpot server is maintained by HubSpot. For open-source alternatives, review the code before connecting it to your CRM data.
• Monitor API usage. Check your Private App's usage dashboard in HubSpot Settings to track what the MCP server is doing.
• Consider a sandbox first. HubSpot offers developer test accounts — test your MCP setup there before connecting to production.

Related guides

• What Is an MCP Server? — start here if MCP is new to you
• Postgres MCP Server — connect AI to your database for direct queries
• Snowflake MCP Server — AI-powered data warehouse analytics