HubSpot and MCP fit together in two directions. One connects an outside assistant like Claude to your CRM, the direction most guides cover.
The other is specific to Breeze, where HubSpot's own agents reach out to external MCP servers and pull data from the systems your team already runs.
This guide separates the two and walks through connecting an external server to a Breeze agent. It also covers where ZoomInfo fits, since HubSpot chose it to power the Breeze Prospecting Agent, for the RevOps and marketing ops teams deciding what to connect Breeze to.
What Is HubSpot Breeze MCP
Breeze is HubSpot's AI brand, spanning the Breeze assistant, the prebuilt agents, and the custom agents you build in HubSpot.
Model Context Protocol is the open standard, created by Anthropic, that lets AI tools connect to external data and tools through one interface. Breeze MCP is where the two meet.
HubSpot added the client-side capability in late 2025 and has expanded its connector set since, and it runs in two directions that get mixed up constantly.
Direction | What it does | Who calls whom | Example |
HubSpot MCP Server | Exposes CRM data to outside AI clients | External client calls HubSpot | Claude reads a deal's history from your portal |
HubSpot MCP Client (Breeze) | Lets Breeze agents pull from external MCP servers | Breeze agent calls an external tool | A Breeze agent analyzes calls in Gong mid-workflow |
The server direction is the more familiar one. Its setup, the CRM objects it exposes, and how it compares to the traditional API are covered in the HubSpot MCP guide. What follows is the Breeze-specific direction, where the agent makes the calls.
How the HubSpot MCP Server Works
In the server direction, HubSpot hosts an MCP server that exposes your CRM to any MCP-compatible client. You authorize it with OAuth, and an assistant like Claude or Cursor can then read and write contacts, companies, deals, and tickets in plain language.
The detail that matters at setup is scope. The connection inherits the permissions of the user who authorized it, so an admin's token exposes far more than a single rep's. Grant the narrowest access the job needs.
This is the path when someone is working in an outside assistant and wants CRM data in the conversation, pulling a record for pre-call research, moving a deal stage, or rolling up pipeline without building a report.
How to Use Breeze as an MCP Client
In the client direction, a Breeze agent connects to an external MCP server and calls it as a tool while it runs.
HubSpot supports a curated set of servers rather than arbitrary ones, currently Notion, Atlassian, Asana, Zapier, G2, Linear, Gong, and Amplitude, so an agent can reach call intelligence, project tools, review data, and product analytics without custom code.
Connecting one to an agent follows a set flow.
Clone the agent. In Agent Hub, open the Agents tab, hover the agent you want to extend, and clone it. HubSpot has you work on a clone rather than the original.
Add the connector. In the cloned agent's Actions, click Add action, open the Connectors tab, choose the MCP server, and authorize it. Most connectors use OAuth, while Zapier uses a tokenized URL you paste in.
Write the prompt instructions. Tell the agent which connected server to use, what to retrieve, and when. The agent will not call an external tool well without explicit direction on when to reach for it.
Writing prompt instructions for Breeze MCP tools
The connection is the easy part. Whether the agent uses the tool correctly comes down to how you write its instructions, and HubSpot's own guidance points to a few practices that carry most of the weight.
Name the source. State which server and which specific dataset the agent should use, and include an identifier where the platform has one, so the agent does not guess.
Describe the structure. List the fields or properties the agent will work with and what each one holds, including the ones that are optional.
Define the task precisely. Say exactly what to find or extract, using concrete names, statuses, and time windows rather than vague asks.
Specify the tool. When an agent has more than one tool available, name the exact action it should call so it does not pick the wrong one.
Give an example input and output. Show what a good result looks like, and set defaults for any fields the agent writes back.
A Breeze account-intelligence workflow in practice
HubSpot's own example connects two servers to one agent. You point the agent at Gong to analyze recent calls for competitive objections, and at Notion to store what it finds. The instructions tell the agent to review calls for each account, pull out competitor mentions and their context, and log a short summary to a named Notion database in a set format.
The result is account intelligence that assembles itself. Reps stop manually reviewing recordings and searching for competitor mentions across systems, and the agent produces a consistent summary they can act on. The same setup works with any server on the list, not just Gong and Notion.
Choosing servers for your own AI client rather than Breeze? Our roundups of the best MCP servers for sales teams and best MCP servers for RevOps compare the field.
Limitations of HubSpot Breeze MCP
HubSpot Breeze MCP has a handful of hard limits worth knowing before you build, and hitting one mid-setup is the usual reason things stall.
You are limited to HubSpot's connector list. If the tool you need is not one HubSpot supports, you cannot point an agent at it directly. Zapier can sometimes bridge the gap, otherwise you wait for HubSpot to add it.
What a connector exposes varies. Each one surfaces only certain actions, so an agent might read from a tool but not write back to it, or reach part of what the tool does rather than all of it.
The agent will not figure the tool out on its own. A connected server does nothing until your prompt instructions tell the agent when to call it, so a vague brief means the tool gets used inconsistently or ignored.
MCP expands reach, not permissions. A connector gives the agent nothing it could not already access in that tool, since it respects the tool's own scopes and your subscription to it.
It connects tools, it does not supply data. Breeze MCP links agents to systems you already run. It does not add verified external contacts or companies, which is a separate layer, and the reason HubSpot pairs the Prospecting Agent with ZoomInfo.
What HubSpot Breeze MCP Costs
HubSpot Breeze MCP has no license fee of its own. What you actually pay for falls into a few parts.
Connectors are free to add. There is no fee to connect a supported MCP server to an agent.
Availability follows your plan. HubSpot MCP Client comes with the Starter, Professional, and Enterprise tiers of HubSpot's hubs, and Smart CRM at Professional and Enterprise.
Some agent runs use HubSpot Credits. HubSpot sells credits in packs, at around $9 per 1,000 on annual billing.
Connected tools bill separately. Any third-party server you attach needs its own subscription.
The separate Breeze Intelligence enrichment product runs on the same credit system and has its own pricing breakdown. On the ZoomInfo side, it is free to start with consumption credits based on usage, and connecting it to Breeze adds no separate MCP fee.
How ZoomInfo Powers the Breeze Prospecting Agent
ZoomInfo is not one of the connectors you add in Agent Hub, because it works with Breeze at a different level.
ZoomInfo is an all-in-one AI GTM Platform built as three stacked layers. A Data foundation of verified B2B contacts and companies sits at the base, the GTM Context Graph reasons over it alongside your CRM and conversation signals, and Universal Access puts it to work in any tool, including a headless API and MCP option for agents.
That access layer is how ZoomInfo reaches Breeze. HubSpot chose ZoomInfo to power its Breeze Prospecting Agent, so the agent builds its lists from ZoomInfo's verified data and Context Graph reasoning rather than scraped or stale records. HubSpot users with a ZoomInfo account turn it on in the Prospecting Agent's settings.
That grounding lets the agent do things it cannot do from CRM data alone.
Build the buying committee. Run contact enrichment to return the right people with verified business emails and direct dials.
Add company context. Layer in firmographics, technographics, and corporate hierarchy.
Surface timing. Pull intent, funding, and leadership changes that show which accounts are in-market.
Keep outreach deliverable. Run email verification so messages land instead of bouncing.
Work from current data. Draw on 600M+ contacts and 100M+ companies, continuously refreshed.
Beyond the prospecting agent, the ZoomInfo MCP server brings the same verified data into other AI tools your team uses, including Claude, ChatGPT, Microsoft Copilot, and Perplexity, all querying one governed source. It is read-only by design, so to keep CRM records themselves current, ZoomInfo enriches HubSpot and Salesforce through its native integrations or the ZoomInfo API, using identity resolution to match a partial record to the right account.
ZoomInfo has a short walkthrough of connecting Claude to ZoomInfo MCP end to end.
How Breeze MCP and ZoomInfo Work Together
Breeze MCP is the connector layer that lets your agents act across your workflow apps. ZoomInfo is the verified-data layer those agents reason over, and it already powers Breeze's own Prospecting Agent. They do different jobs, so most teams end up using both.
| HubSpot Breeze MCP | ZoomInfo on GTM AI |
What it is | Connectors that link Breeze agents to external tools | An AI GTM Platform: verified data plus Context Graph intelligence |
Connects to | A curated set such as Notion, Asana, Gong, and G2 | Any MCP client, plus HubSpot's Breeze Prospecting Agent |
Data source | The workflow apps you connect | 600M+ contacts and 100M+ companies, continuously refreshed |
Best used for | Letting an agent act across your stack | Grounding agents in data the CRM does not hold |
On its own, a Breeze agent can only reason over what it can reach, so a thin or decayed record produces a thin answer. Backed by ZoomInfo, the same agent verifies a contact before it drafts outreach and pulls intent before it prioritizes an account. One layer supplies the reach across your tools, the other supplies the data it acts on.
For vendor-by-vendor views of how the data layers compare, see Clay MCP and Apollo MCP.
Ground Your Breeze Agents in Verified Data
Breeze MCP lets your HubSpot agents act across your stack, and the data underneath decides whether they act on something true.
HubSpot already chose ZoomInfo to power its Breeze Prospecting Agent, and the same verified data reaches your other agents through ZoomInfo MCP. Start building on GTM.AI to put verified data behind every agent.
Frequently Asked Questions
Do I need a developer to connect an MCP server to a Breeze agent?
No. The connection happens in Agent Hub without code, and most servers authorize with a login. The skill that matters is writing clear prompt instructions, which is a writing task more than an engineering one.
Why isn't my Breeze agent using a connected tool?
It is usually one of three things: the prompt instructions do not tell the agent when to call the tool, you edited the original agent instead of the clone, or the clone you configured is not the version running. Check those before anything deeper.
Does connecting a tool give it access to my HubSpot data?
No. In the client direction the agent reaches out to the tool and sends only what its task needs, so it is not exposing your CRM. Sharing CRM data outward is the server direction, a separate connection you authorize on its own.
Can I connect ZoomInfo to HubSpot Breeze?
Not as a Breeze MCP connector, since ZoomInfo is not on HubSpot's supported list. HubSpot uses ZoomInfo to power its Breeze Prospecting Agent, grounding it in verified data and Context Graph intelligence, and ZoomInfo MCP brings the same data into other AI clients like Claude and ChatGPT.
