HubSpot MCP server

Connect Claude and ChatGPT to HubSpot without losing control of your data

HubSpot's MCP server lets AI tools read and update your CRM as if they were the person signed in. HubSpot's connectors for Claude, ChatGPT, Microsoft Copilot and Gemini all run on it. We set it up with the right permissions and build custom agents on top of it.

Free, 30 minutes, and you'll leave with at least one clear next step.

Hub
Any HubSpot account (Claude connector: all products and plans)

AI tools act with each user's own HubSpot permissions

How does the HubSpot MCP server connect AI tools to your CRM?

MCP stands for Model Context Protocol. It's a standard way for AI tools to talk to other software. HubSpot runs one MCP server for this, and every AI tool on the left goes through it.

CRM

AI tools

  • Claude
  • ChatGPT
  • Microsoft Copilot
  • Gemini
  • Custom agents you build

Through mcp.hubspot.com

  • A question or update, signed in as the user CRM to ERP
  • Only the records that user is allowed to see ERP to CRM
ERP

Your HubSpot account

  • Contacts, companies and deals
  • Tickets and any custom record types
  • Calls, meetings, notes and tasks
  • Campaigns, emails and pages
  • Products and line items
Sign-in uses each person's own HubSpot login, so the AI tool can't see more than they can.

What can you do with it?

Your team can ask an AI tool questions about your CRM and have it make updates, without exporting spreadsheets. A sales manager can ask Claude which large open deals haven't had a call this month. A rep can paste meeting notes into ChatGPT and have it log the call, update the deal and create follow-up tasks.

The same server also works for custom agents. So an agent that reads an emailed request, finds the account in your CRM and creates the deal can use the same permission-limited connection your team uses in Claude.

What can the HubSpot connector for Claude read and write?

HubSpot's help pages spell out what its Claude connector can do. It works on every HubSpot product and plan, and it needs a paid Claude plan.

  • Read, create and update contacts, companies, deals, tickets and other CRM records.
  • It can't delete anything.
  • Users and teams are read only, so it can't change who has access.
  • Bulk changes are capped at 10 records at a time, so large updates still need an import or sync.
  • Each Claude account connects to one HubSpot account at a time.
The problems we see

What should you set up before your team connects?

Each person's AI tool gets that person's HubSpot permissions. That's safe when permissions are tight. If they're loose in HubSpot, they're loose in the AI tool too.

Tighten HubSpot permissions first

If every user can see every deal today, every user's AI tool can too. We review roles and teams before anyone connects.

Decide who can install it

Super Admins and users with App Marketplace access can connect without approval. Everyone else needs a Super Admin to approve it, and the admin picks which data the connector can reach.

Know the limits

Bulk creates and updates are capped at 10 records at a time, and everything runs within HubSpot's usual usage limits. Large updates still need a data import or sync.

Plan for sensitive data

If your HubSpot account has its Sensitive Data setting turned on, Claude can't see activity history like calls and emails.

Should you use an AI connector, Agent Builder or a custom agent?

There are three ways to put AI on top of your HubSpot data. Most companies end up using more than one.

AI connector or Agent Builder

  • A person asks, the AI answers or updates (connector)
  • Agents that live inside HubSpot and use credits (Agent Builder)
  • Agent Builder agents can also reach some outside tools
  • Fast to start, little to maintain
  • Best for one-off questions and HubSpot-only jobs

Custom agent on the MCP server or API

  • Runs on its own, on a schedule or trigger
  • Reads PDFs, accounting or back-office data, or other systems HubSpot can't
  • You pick the AI model and control the prompts
  • Business rules and approval steps you define
  • Best for repeat jobs at volume
How it works

Permissions come first, then connections

You provide a Super Admin, a list of who should use AI tools, and the jobs they want help with. You get a connected team with sensible access and, where it pays, a custom agent.

  1. Audit permissions

    We check roles, teams and who can see which records in HubSpot, so each person's AI tool sees only what it should.

  2. Approve and connect

    We set which data the connector can reach and who can install it, then connect your first users in Claude, ChatGPT, Copilot or Gemini.

  3. Train on everyday tasks

    We show your team prompts for the work they do every day, like prepping for a call or logging notes, using your own records.

  4. Find the repeat jobs

    After a few weeks, we look at what people keep asking for by hand. Those are the jobs worth turning into an agent.

  5. Build the custom agent

    We build agents in Agent Builder when the job fits inside HubSpot, or outside HubSpot when it needs your other systems, documents or your own rules.

When should a job run without anyone asking?

A connector works well when a person asks the questions. Some jobs should happen on their own, every time a request comes in or every night from another system's data. Those are custom AI workflows built on the MCP server or HubSpot's API, with retries and approval steps. We've connected HubSpot to other systems for clients like Photon Brothers, which saves about 10 hours of manual data entry a week. For syncing with an ERP or back-office system, see our guide to connecting HubSpot to your ERP.

FAQ

MCP server questions

What is the HubSpot MCP server?

It's a HubSpot service that lets AI tools read, create and update HubSpot records the way the signed-in user would. HubSpot's connectors for Claude, ChatGPT, Microsoft Copilot and Gemini run on it, and custom agents can use it too.

Is it safe to connect ChatGPT or Claude to our CRM?

The AI tool signs in as the person using it and gets only that person's HubSpot permissions. So it's as safe as your permissions are. We review roles and data access before anyone connects.

Do we need a paid Claude account to connect it?

Yes. HubSpot's Claude connector needs a paid Claude plan, such as Pro, Max, Team or Enterprise. On the HubSpot side, it works with every product and plan.

Can the AI delete records?

Not through HubSpot's Claude connector. It can read, create and update records, but it doesn't support deleting them.

Is there a separate MCP server for developers?

Yes. HubSpot also has a Developer MCP server for people building apps and websites on HubSpot. The main MCP server is the one for working with CRM data.

Can HubSpot's own agents use outside MCP servers?

Yes. Agents made in Agent Builder can call supported outside MCP servers, so a HubSpot agent can reach tools beyond HubSpot.

When should we build a custom agent instead?

When the job repeats often, runs without a person asking, or needs data from documents or other systems. A connector is best for people asking questions. A custom agent is best for work that should happen on its own.

Connect your AI tools to HubSpot the right way

On a call we'll look at your HubSpot permissions and the jobs your team wants AI help with, and tell you whether a connector is enough or a custom agent is worth building.