AI and HubSpot: Automating Your CRM the Right Way

HubSpot is one of the most widely used CRMs among small and mid-sized businesses, and one of the first vendors to publicly document both a built-in AI layer and an open connection for external agents. This article sets out what actually exists today, without overselling the vendor's own numbers. It builds on our overview of AI for omnichannel customer service.

What is Breeze, HubSpot's native AI layer?

Breeze is the name HubSpot gives to the set of AI agents built directly into the CRM. The vendor lists several on its official product page: a Prospecting Agent, a Customer Agent, a Data Agent, plus an Agent Builder for creating custom agents and a Smart Deal Progression mechanism that moves certain deals forward automatically according to rules you define. Source: hubspot.com/products/artificial-intelligence.

The common thread across these agents: each targets a specific, repetitive task in day-to-day sales or support work, rather than a general-purpose assistant that does a bit of everything.

What do the Prospecting Agent and the Customer Agent actually do?

The Prospecting Agent qualifies and prioritises inbound leads in place of a sales rep manually sorting a list. HubSpot reports an average 65% increase in leads processed per month for teams using it, a figure to read as a vendor average rather than a guarantee that transfers unchanged to every organisation.

The Customer Agent resolves some customer service tickets on its own. HubSpot cites an average 77% increase in tickets closed per month. Here too, the result depends heavily on the volume of repetitive tickets already present and the quality of the knowledge base the agent draws on.

What does the Data Agent do?

The Data Agent speeds up research on a prospect or client by pulling from data already in the CRM and from external sources. HubSpot claims research up to 10 times faster than manually cross-referencing several tabs and tools. It is a time saver on a preparatory task, not an automated decision on the account itself.

How does an external AI agent connect to HubSpot?

Beyond Breeze, HubSpot officially documents two MCP servers, the open standard that lets an agent (Claude, ChatGPT or an in-house build) read and write to a tool without a bespoke integration built from scratch for every case. Source: developers.hubspot.com/mcp.

The Remote CRM MCP (mcp.hubspot.com) gives read and write access across the whole CRM through OAuth authentication. An external agent can query the pipeline in plain language, for example "which deals this month have had no follow-up in ten days", or update contact properties according to rules you define.

The Developer MCP runs locally and is aimed at building applications or customisations on the HubSpot CMS, a use case closer to a developer's workflow than a day-to-day sales one.

On the exact general-availability dates for both servers, and on any data-volume threshold for some of the CRM's predictive scores, only third-party sources are circulating at the time of writing: check the current state on the official documentation before building a project on top of it.

How do you combine Breeze with an external agent connected over MCP?

The two approaches are not competing. Breeze fits well with tasks HubSpot has already packaged: qualifying inbound leads, first-level ticket handling, prospect research. An external agent connected to the Remote CRM MCP takes over on needs specific to the business that do not fit the vendor's pre-built scope: cross-referencing the HubSpot pipeline with production or stock data, orchestrating a follow-up that follows a business-specific logic, or querying the CRM from an agent already used for other tools.

This shared-connection-layer logic across several systems is the same one described for unifying customer service channels beyond the CRM alone: the CRM is just one of the sources the agent needs to consult, not the only one.

What guardrails should you set before connecting an agent to your CRM?

The first instinct is to limit write access. The Remote CRM MCP allows editing records and deals: start in read-only mode (reporting, research, pipeline summaries) before opening up write access, reserved for reversible, low-risk actions.

The second guardrail concerns the personal data held in a CRM: contact details, exchange history, sometimes sensitive information depending on the sector. The CNIL points out that any AI processing of personal data must stay proportionate and documented, with a scope defined in advance, which applies to a CRM the same as any other business tool.

The third point is traceability: every action the agent takes, whether from Breeze or an external agent over MCP, needs to be traceable after the fact, especially for bulk changes.

Where do you start with AI and HubSpot?

The safest sequence is the same as for any CRM connected to an agent: start with a read-only use that changes nothing (research, reporting), before gradually opening up write access on precise, reversible tasks. The Prospecting Agent, already packaged by HubSpot, is often the fastest entry point for a sales team, ahead of considering an external agent connected to the Remote CRM MCP for needs more specific to the business.