Short answer: HubSpot MCP is an official server built on the Model Context Protocol that lets outside AI assistants, including Claude, ChatGPT, Gemini and Microsoft Copilot, connect to your HubSpot CRM and query real data in plain language. It is included with a HubSpot account, does not require code, and is different from Breeze Agents, which are HubSpot's own AI running inside the platform. Start read-only, on one object type, with one team, then expand once you trust what it returns.
For the last two years, the standard way to get AI help with CRM data was to export a report, paste it into a chat window, and ask a question about numbers that were already a day old. That workaround is going away. HubSpot has shipped an official MCP server, and the major AI assistants your team already uses can now connect to it directly, which means the question changes from “how do I get this data into ChatGPT” to “how much of my CRM should I actually let ChatGPT see.”
This matters more for SMBs than for enterprises with a data team, because most SMBs on HubSpot do not have anyone whose job is to build a custom integration. MCP is designed to be turned on in the settings panel, not built by a developer. That accessibility is the opportunity and the risk in the same feature.
01. What HubSpot MCP Actually Is
MCP stands for Model Context Protocol, an open standard originally published by Anthropic that defines a common way for AI assistants to connect to external tools and data sources. Instead of every company building a separate, one-off integration for every AI assistant, a service publishes one MCP server, and any MCP-compatible assistant can connect to it using the same protocol. HubSpot has published its own official MCP server, and the major AI assistants, Claude, ChatGPT, Gemini and Microsoft Copilot, have each added support for connecting to remote MCP servers like it.
In practice, this means a sales manager can open Claude or ChatGPT, connect it to HubSpot once, and then ask a question like “which deals in the manufacturing segment have had no activity in 14 days” and get an answer pulled from live CRM data, not from a spreadsheet someone exported last Tuesday. The AI assistant is not guessing. It is querying HubSpot through the connector and reasoning over the actual records it gets back.
Why it matters: this closes the gap between where your team already works, inside Claude, ChatGPT or Copilot, and where your customer data actually lives. The alternative for most SMBs was either no connection at all or a fragile, manually maintained export, and MCP replaces both with a supported, native path.
02. Three Ways to Put AI on Top of HubSpot
MCP is not the only option, and it is easy to confuse it with tools that solve a similar-sounding problem. These are the three practical paths and where each one fits.
| Approach | Who builds and runs it | Best for | Setup effort |
|---|---|---|---|
| HubSpot MCP / AI Connectors | HubSpot's own server, connected from Claude, ChatGPT, Gemini or Copilot | Plain-language questions and drafting, asked from an AI assistant your team already uses | Low, no code, settings-panel connection |
| Breeze Agents | HubSpot, running autonomously inside the CRM | Ongoing tasks like ticket deflection, prospecting research and data cleanup | Low, but tier and credit dependent |
| Custom API or Zapier / Make automation | You or a developer, calling HubSpot's API or a workflow tool | Fixed, repeatable automations not tied to a chat interface | Medium to high, ongoing maintenance |
These are not mutually exclusive. A well-run HubSpot portal in 2026 often uses all three: Breeze for tasks that should run on their own inside HubSpot, MCP for ad hoc questions asked from whichever AI assistant a person prefers, and automation tools for fixed workflows that do not need a conversation at all.
Why it matters: teams that treat MCP, Breeze and automation as interchangeable end up either overbuilding a custom integration for a question an AI connector already answers, or asking an AI assistant to do a fixed, repeatable job that a simple workflow would run more reliably and more cheaply.
03. What You Can Actually Ask a Connected AI Assistant
Once the connection is live and scoped, the questions that get the most immediate value are the ones that used to require someone to build a report first.
- Pipeline and revenue questions in plain language.“What is our open pipeline in the wholesale segment this quarter” without opening a dashboard.
- Stalled deal and overdue task flags.“Which deals over $10,000 have had no logged activity in two weeks” surfaced instantly instead of during a manual pipeline review.
- Drafting with real context. A follow-up email drafted using the actual deal history and last contact note, not a generic template.
- Account and contact lookups mid-conversation.A rep checking a company's order history and open tickets without switching tabs.
With write access enabled and scoped carefully, some teams go further and let the assistant update a property or log an activity directly, though most SMBs are better served keeping the first months of use read-only while trust and habits build.
Why it matters:the value is not a new capability so much as removed friction. None of these questions were impossible before. They required someone to build a report, run an export, or interrupt a rep's day. MCP moves the answer to wherever the question was already being asked.
04. Setting Up the Connection
The setup itself is a settings-panel task, not a development project, though the decisions around scope deserve real attention before you flip it on.
- 1. Decide the scope before you connect anything. Pick the objects (deals, companies, contacts, tickets) and the read or write level a first pilot actually needs. Narrower is the right starting point.
- 2. Enable the connector in HubSpot. In your HubSpot account's AI or integrations settings, turn on the MCP server or AI Connectors feature and generate the authenticated connection for the scope you chose.
- 3. Add HubSpot as a connector in your AI assistant. In Claude, ChatGPT, Gemini or Copilot, follow that tool's process for adding a remote connector or MCP server, and authenticate using the credentials HubSpot generated.
- 4. Pilot with one team for two to three weeks. Give access to the people who will actually use it daily, and keep the group small enough that you can review what was asked and what was returned.
- 5. Review the audit log, then expand. Check HubSpot's connector activity log for what was queried and whether the answers were accurate before widening scope, objects or user count.
Why it matters: because the connection is genuinely easy to enable, the discipline has to come from the sequence, not from the setup difficulty. A five-minute setup with no scoping plan is how a narrow, useful pilot turns into an unmonitored connection touching every object in the portal.
05. Security, Permissions and Data Governance
Connecting an outside AI assistant to your CRM raises the same question every integration raises: what happens to the data once it leaves HubSpot, and who can see it.
Scope by object and property, not by user role alone.
A connection scoped to deal amounts and stages is a very different risk than one scoped to every contact property including notes and personal details. Grant the narrowest scope that answers the questions your pilot team actually needs answered.
Read-only first, write access reviewed and deliberate.
An AI assistant that can only read data cannot change or delete a record by mistake. Reserve write access for a later phase, a smaller group, and a specific, well-tested use case.
Know where the AI vendor processes data.
Under PIPEDA and provincial privacy expectations, a Canadian SMB should be able to state where customer data goes once an AI assistant queries it, whether that data is retained by the AI vendor, and how a customer's access or deletion request would be honoured across both systems. Confirm this before rolling a connector out beyond a small pilot.
The governance rule that holds up:
If you would not put a piece of data in an email to a stranger, do not put it in the scope of an AI connector without a specific reason. Most CRM data does not carry that risk. Some of it, like payment details, health information or sensitive personal notes, does, and it should stay out of scope unless there is a clear, reviewed business reason to include it.
Why it matters: the reputational and compliance cost of an overexposed connector lands well before the cost of a delayed AI rollout. Scoping deliberately from day one is cheaper than explaining an incident after the fact.
06. What This Means for Canadian SMBs
Canadian SMBs on HubSpot tend to run lean, without a dedicated RevOps or data team, which is exactly the profile MCP is built for. A settings-panel connection that does not require a developer removes the usual barrier that kept AI-and-CRM projects on a roadmap instead of in production.
For manufacturers and wholesalers running Shopify B2B alongside HubSpot, the same connector can answer questions that currently require checking two systems, such as a wholesale account's order history in Shopify next to its deal stage in HubSpot, once both are represented in the CRM record. Bilingual operations in New Brunswick and Quebec should also confirm that a customer-facing use of a connected assistant handles French output correctly before relying on it for anything client-visible.
The teams already running HubSpot Breeze agents have a head start here, because the same data discipline, clean records and a documented process, that makes Breeze agents safe also makes an MCP connection trustworthy. If your foundation is not there yet, our guide to HubSpot integrations and automations for B2B covers the sequencing that should come first.
Why it matters: being small is an advantage here, not a limitation. A lean team can pilot, review and expand an MCP connection in weeks, while a larger organization often needs a security review cycle to do the same thing.
07. Where MCP Fits Into Your AEO Strategy
There is a second, less obvious angle to HubSpot MCP: it is part of the same shift toward AI-mediated interactions that answer engine optimization is built around. If buyers are increasingly asking Claude or ChatGPT questions and getting answers pulled from structured sources rather than clicking through search results, then a business that has already made its own data legible to AI assistants, internally through MCP and externally through structured, answer-first content, is better positioned on both fronts.
Internally, that means CRM data an AI assistant can query cleanly because properties are consistent and records are deduplicated. Externally, it means the same discipline applied to your published content, which our guide to GEO and AEO for Canadian ecommerce covers in more depth. The two are related applications of the same underlying skill: making your information something an AI system can read and trust.
Why it matters: the businesses treating AI-readability as a single project, covering both internal data and external content, will spend less time retrofitting each new AI feature that shows up in 2026 and beyond.
08. Key Takeaways
- HubSpot MCP is an official server built on the Model Context Protocol that lets Claude, ChatGPT, Gemini and Copilot query your CRM data directly, without a custom integration.
- It is different from Breeze Agents, which are HubSpot's own AI running inside the platform. MCP is you bringing an outside AI assistant to your HubSpot data.
- Setup is a settings-panel task on both sides, HubSpot and the AI assistant, with no code required for a basic connection.
- Start read-only, on a narrow set of objects, with one team, and review the connector activity log before expanding scope or granting write access.
- Canadian SMBs should confirm where the connected AI vendor processes data before rolling a connector out beyond a small pilot, given PIPEDA expectations.
- Lean teams have a real advantage here: a small, well-scoped pilot can be running in a single afternoon.
09. Frequently Asked Questions
What is HubSpot MCP?
HubSpot MCP is HubSpot's official server built on the Model Context Protocol, an open standard for connecting AI assistants to external data sources and tools. It lets AI systems such as Claude, ChatGPT, Gemini and Microsoft Copilot query and, with the right permissions, update your HubSpot CRM data directly, instead of you copying deal or contact information into a chat window by hand. It sits alongside HubSpot's own Breeze AI agents as a second, more open path for putting AI on top of your CRM.
How do I connect HubSpot to ChatGPT or Claude?
In HubSpot, go to the AI Connectors or Integrations settings and enable the MCP server for your portal, which generates an authenticated connection. In Claude, ChatGPT or another MCP-compatible client, add HubSpot as a connector using that authentication, following the app's own steps for adding a remote MCP server or connector. Once connected, you can ask questions about your CRM data in plain language and the assistant queries HubSpot in real time, scoped to whatever permissions you granted the connection.
Is HubSpot's MCP server free?
The HubSpot MCP server itself is included with a HubSpot account and does not carry a separate HubSpot fee at launch. The cost sits on the other side of the connection: Claude, ChatGPT, Gemini and Copilot each have their own subscription and usage pricing, and heavier query volume through an AI assistant can draw on that tool's own API or seat costs. Budget for the AI assistant's plan, not for HubSpot MCP as a line item.
What can an AI agent actually do inside HubSpot through MCP?
Depending on the scopes you grant, a connected AI assistant can search and summarize contacts, companies and deals, pull pipeline and revenue figures into a plain-language answer, draft follow-up emails using real deal context, flag stalled deals or overdue tasks, and in some configurations create or update records. Read-only access is the safer starting point for most SMBs. Write access should be scoped narrowly and reviewed before it is turned on broadly.
Is it safe to connect Claude or ChatGPT to my CRM data?
It is safe when the connection is scoped deliberately: start with read-only access, limit which objects and properties are exposed, restrict the connection to a small group of users, and review HubSpot's audit log for connector activity. It is not safe to grant broad read-write access to every object on day one, because an AI assistant acting on a vague instruction can update or delete records at a speed a person would not. For Canadian SMBs, also confirm where the connected AI vendor processes data, since that matters for PIPEDA and customer-facing privacy notices.
What is the difference between HubSpot MCP and Breeze Agents?
Breeze Agents are HubSpot's own AI agents, built and run inside HubSpot to handle tasks like prospecting, support and data cleanup autonomously. HubSpot MCP is a connector that lets AI assistants built by other companies, such as Claude, ChatGPT, Gemini or Copilot, reach into your HubSpot data from outside the platform. Breeze is HubSpot doing the AI work for you inside its own product. MCP is you bringing your own AI assistant to your HubSpot data. Many teams end up using both for different jobs.
How should a Canadian SMB get started with HubSpot MCP?
Start with one connector, one AI assistant, and read-only access to a narrow set of objects, such as deals and companies for a sales manager who wants plain-language pipeline answers. Confirm with your team which questions the connection is meant to answer, test it for a couple of weeks, and check the audit log for what was actually queried. Only expand scope, users or write access once that pilot shows the connection is used, accurate and governed, the same sequencing that applies to any AI rollout on top of a CRM.
Related Resources
What each Breeze agent does, what it costs, and how to deploy it safely
What a HubSpot partner does, every Hub explained, and what to ask before you sign
The integrations and automations to build first, and in what order
How to get cited in AI-generated answers, not just ranked in search
Thinking about connecting AI assistants to your HubSpot data?
AtlanticWorks implements HubSpot for Canadian SMBs and manufacturers, including the data cleanup, permission scoping, and connector pilots that make MCP and AI Connectors safe to turn on. The free assessment shows where your portal stands and what to fix before you connect anything.
Start the Assessment