What Is MCP (Model Context Protocol)?
MCP (Model Context Protocol) is a standard way to connect an AI assistant to the information and tools in another system, such as a CRM. The assistant handles the request; MCP gives it a controlled way to reach the supporting data and actions. What the assistant can do depends on the connection and the permissions it has been given.
Key takeaways
- MCP is a connection standard, not an AI product. It lets assistants like Claude, ChatGPT and Copilot read from and act in systems like HubSpot.
- The assistant does the thinking; the connected system does the work. MCP is the agreed way they talk to each other.
- Permissions decide everything. An MCP connection can be read-only, or allowed to create and update records, and that choice should be deliberate.
MCP explained without the jargon
Think of an AI assistant as a capable new colleague who has never seen your systems. They can reason and write well, but they cannot look anything up or change anything on their own. MCP is the badge and the key card: it tells the assistant what it is allowed to see and do in a given system, and gives it a standard way to ask.
Before MCP, every AI tool needed its own custom integration with every system. With MCP, a system exposes one connection that any compatible assistant can use, and the assistant does not need to know how HubSpot, a file store or a calendar works internally.
A worked example
A marketer asks their assistant: "Find everyone who attended yesterday's webinar and draft a follow-up email with the recording link." Through an MCP connection to HubSpot, the assistant retrieves the attendee list, pulls the recording link from the connected content system, and prepares a draft for review. The person still decides whether to send it.
The assistant handled the request and the writing. MCP provided the route to the attendee data and the recording. Nothing was sent without a human looking at it.
MCP in a HubSpot context
HubSpot operates its own MCP server, which is what the ChatGPT, Claude and Microsoft Copilot connectors use to work with a portal. HubSpot's Fall 2026 developer release expanded the set of actions available to external AI tools through that server, so an assistant can do more than read: depending on permissions it can search records, summarize deals, draft emails, or prepare updates for approval.
In practice the connection is only as good as the portal behind it. If lifecycle stages are inconsistent, properties are half-filled or ownership is unclear, the assistant inherits those problems and produces confident answers built on bad data. Clean definitions, scoped permissions and a named owner for the connection matter more than the choice of assistant.
| What the team wants | What MCP makes possible | What still needs a person |
|---|---|---|
| A summary of an account before a call | Assistant reads the company record, open deals, recent activity and notes | Deciding what to ask and what to promise |
| Follow-up after a webinar or event | Assistant pulls attendees and prepares personalized drafts | Reviewing tone, checking consent, pressing send |
| Keeping deal data current | Assistant proposes property updates from meeting notes | Approving changes before they land in the CRM |
| Answering "which deals slipped this month?" | Assistant queries deals and returns a list with reasons where recorded | Interpreting why, and what to do about it |
Questions to ask before switching it on
- Which assistant will people actually use, and does it have a maintained connector for HubSpot?
- Should the connection be read-only to start, and which record types should it see?
- Who owns the connection, reviews what it does and turns it off if something goes wrong?
- Which one process do we test first, and how will we know whether it helped?
Frequently asked questions
What does MCP stand for?
Model Context Protocol. It is an open standard for connecting AI assistants to external data and tools in a consistent way.
Is MCP a HubSpot feature?
No. MCP is an industry standard. HubSpot supports it by running an MCP server that assistants such as Claude, ChatGPT and Microsoft Copilot connect to, and by shipping connectors built on it.
Can an assistant change data in HubSpot through MCP?
Only if the connection and the user's HubSpot permissions allow it. Many teams start read-only and add create or update actions once they have seen how the assistant behaves.
Do we need developers to use MCP with HubSpot?
Not for the standard connectors, which a Super Admin can enable. Custom integrations or extra actions beyond what HubSpot exposes do need technical work.
Is MCP the same as HubSpot Breeze?
No. Breeze is HubSpot's own AI inside the portal. MCP is how outside assistants reach HubSpot data. Many teams use both.
Related service: Set up HubSpot MCP with Superwork