Last verified: June 2026
There are two AI stories happening inside HubSpot right now.
The first is Breeze, HubSpot’s own AI built into the platform. The second is that HubSpot now lets you pull your entire CRM into the AI tools your team already uses: Claude, ChatGPT, Gemini, or Microsoft Copilot. You can ask your pipeline questions in plain English without ever opening a report builder. The first story is about HubSpot getting smarter. The second is about HubSpot becoming the brain behind every AI conversation your team has.
This is the field guide to both, written for the RevOps leader who is tired of AI hype and wants to know what is real, what is worth turning on, and what to fix before any of it can help. Because the uncomfortable truth underneath all of this is simple: AI on top of bad CRM data does not give you answers. It gives you wrong answers with more confidence. We will come back to that.
The short version:
- Breeze is HubSpot’s umbrella brand for all of its AI: a free Assistant on every tier, autonomous Agents that now charge per completed outcome, and an Intelligence data layer whose standard enrichment is included with paid seats.
- At the Spring 2026 Spotlight (April 14, 2026), HubSpot shipped 99 updates, headlined by HubSpot AEO, Smart Deal Progression, a rebuilt Prospecting Agent, and official AI Connectors for Claude, ChatGPT, Gemini, and Microsoft Copilot.
- The Remote MCP Server is now generally available, meaning any MCP-compatible AI tool can read and update your standard CRM records, governed by your existing permissions.
- None of it is trustworthy on top of dirty data. Foundation first, AI second.
What is HubSpot Breeze?
What did HubSpot announce at the Spring 2026 Spotlight?
Short answer: Breeze is the umbrella name for every AI capability inside HubSpot. It comes in three distinct forms that do very different things and cost very differently: Breeze Assistant (a conversational helper), Breeze Agents (autonomous workers), and Breeze Intelligence (a data enrichment layer).
Get those three straight and most of the noise clears up. Here is the fast map, then the detail on each.
| Layer | What it does | Availability | Cost in June 2026 |
|---|---|---|---|
| Breeze Assistant | Conversational helper in the sidebar and Slack | Every tier, including free CRM | Free |
| Breeze Agents | Autonomous multi-step workers for support, prospecting, and data | Professional and Enterprise | Outcome-based: $0.50 per resolved conversation (Customer Agent), $1 per recommended lead (Prospecting Agent), paid via credits |
| Breeze Intelligence | Data enrichment and buyer intent | Paid tiers (Starter and up) | Standard enrichment included with Core Seats; Smart Properties and Buyer Intent consume credits |
What is Breeze Assistant, and what does it cost?
Short answer: Breeze Assistant is HubSpot’s conversational AI helper, and it is free on every tier, including the free CRM. There is no upgrade gate.
You may have known it as Copilot, or before that ChatSpot. You ask it something, it answers. Summarize this contact’s history before my call. Draft a follow-up email. Pull up the deals I have closing this month.
The single most underused fact in HubSpot AI is that the Assistant costs nothing, on any plan. If your team is not using it daily, you are leaving the easiest win in the entire platform untouched.
The Spring 2026 Spotlight also made it meaningfully smarter:
- Role-aware responses. The same Assistant now gives a sales rep deal-specific guidance and a marketer campaign-specific guidance, depending on who is asking.
- Trained on Loop Marketing, HubSpot’s playbook for marketing in the AI era. The Assistant can now help define ideal customer profiles, build brand guides, and draft campaign briefs grounded in your actual customer data rather than generic templates.
- Automatic brand alignment on generated content, plus expanded data access that now includes website analytics and HubSpot Academy resources alongside your CRM data.
- Available in Slack, which puts CRM answers in the place many teams already live.
What are Breeze Agents, and how does outcome-based pricing work?
Short answer: Agents are autonomous AI workers available on Professional and Enterprise. As of April 14, 2026, the two flagship agents charge per completed outcome: $0.50 per conversation the Customer Agent actually resolves, and $1 per lead the Prospecting Agent recommends for outreach.
Where the Assistant waits for you to ask, Agents run in the background taking multi-step actions on their own. The main 3 right now:
- Customer Agent resolves support tickets autonomously across live chat, your website, WhatsApp, and (new as of Spring 2026) email. HubSpot reports a 65 percent conversation resolution rate, with resolutions happening 39 percent faster, across roughly 8,000 customers using it.
- Prospecting Agent was rebuilt for Spring 2026. It now researches a target company, identifies likely members of the buying committee, enriches your CRM with those contacts, tracks buying signals such as funding announcements, product launches, and new hires, drafts personalized outreach for rep review, and compiles a pre-call brief once a meeting gets booked.
- Data Agent answers custom business questions by combining your CRM data, conversation transcripts, documents, and the web, and can write its findings back into the CRM.
More agents continue to move through beta on HubSpot’s usual cadence.
The pricing mechanics matter, because they changed in April 2026 and most articles you will find are already out of date. Previously, the Customer Agent charged $1.00 per conversation whether or not it solved anything, and the Prospecting Agent carried a recurring monthly charge per enrolled contact. Both moved to outcome-based pricing effective April 14, 2026. In the words of HubSpot’s chief customer officer Jon Dick: “You pay when it works, full stop.”
Under the hood it still runs on HubSpot Credits. A resolved conversation costs 50 credits, credits run $10 per 1,000 (slightly less on annual commitments), and plans include a monthly allowance: 500 credits on Starter, 3,000 on Professional, 5,000 on Enterprise. Credits do not roll over. Both flagship agents come with a 28-day free trial, which is the right way to test them against your real volume. And because HubSpot has adjusted this model more than once in the past year, confirm the current numbers on HubSpot’s own pricing pages before you forecast spend, including anything you read here.
One more piece of context worth carrying into a buying conversation: HubSpot is not alone in this shift. Intercom, Zendesk, Sierra, and Salesforce have all been moving toward results-based pricing for AI agents. Paying for completed work instead of access is becoming the default economics of agents, and it generally favors the buyer.
What is Breeze Intelligence?
Short answer: Breeze Intelligence is HubSpot’s data enrichment and buyer intent layer, built largely on the Clearbit acquisition. Standard firmographic enrichment is now included free with Core Seats on Starter and above. Only advanced features like Smart Properties and Buyer Intent consume credits.
This is the layer that fills in revenue, industry, employee count, and location on your records, drawing on a database of more than 200 million company and contact profiles. The packaging change that flew under the radar: since HubSpot’s INBOUND 2025 packaging update, that standard enrichment no longer costs credits at all on paid seats. A lot of teams are still budgeting for a cost that quietly disappeared.
What still costs money: Smart Property enrichment (custom AI-researched fields, like hunting down a specific attribute the standard dataset does not carry), Buyer Intent (identifying companies researching your category from anonymous website traffic), and high-volume enrichment. Those run on credit packs ranging from roughly $45 to $700 per month. If you are paying a third party for basic firmographics today, it is worth twenty minutes to check which side of that line your usage actually falls on.
Can you connect HubSpot to Claude, ChatGPT, and Gemini?
Short answer: Yes. HubSpot’s AI Connectors and Remote MCP Server, generally available as of April 2026, let you read and update standard CRM records from Claude, ChatGPT, Gemini, and Microsoft Copilot in natural language, governed by your existing HubSpot permissions.
You can now ask questions of your real customer data inside the AI tool you already use, without copying anything into a chat window or building a report first.
What is MCP?
MCP (Model Context Protocol) is an open standard that lets AI tools securely access and use data from other systems. Think of it as a universal adapter: one protocol that Claude, ChatGPT, and the rest all speak, so any of them can plug into HubSpot through the same connection. Before MCP, your options were manual copy-paste, expensive custom API builds, or limited pre-built connectors. MCP replaces all three with one standardized path.
What does HubSpot’s MCP server actually do?
The official Remote MCP Server and the named AI Connectors are genuinely useful. What they do:
- Connect Claude, ChatGPT, Gemini, and Microsoft Copilot (and any other MCP-compatible client) to your HubSpot portal.
- Provide read, create, and update access to standard CRM objects: contacts, companies, deals, tickets, line items, products, quotes, invoices, orders, subscriptions, and segments. A rep can ask Claude to summarize every deal in the decision-maker stage over $10,000. A manager can pull the five most recent enterprise tickets. All of it comes live from the actual data.
- Respect the connected user’s existing HubSpot permissions. It is not a backdoor. An AI client can only see what that user could already see, and a HubSpot admin controls whether the organization can connect at all.
- Work on every HubSpot tier, including the free CRM, with no separate HubSpot charge.
What are the limitations you should know before rolling it out?
Being clear-eyed about the limits is what separates a real recommendation from a press release:
- No custom objects. The connectors expose standard CRM object types only. If you have modeled subscriptions, usage, or anything custom (and growing SaaS teams usually have), the official server cannot see that data yet. This is the single biggest gap for ops-heavy accounts.
- No delete. Delete functionality is excluded by design to protect data integrity. Read, create, and update only.
- The AI side usually is not free. There is no HubSpot charge, but the one-click connectors generally require a paid plan on the AI platform side (Claude Pro, ChatGPT Plus, Gemini Advanced, or the equivalent). Check current requirements for the tool your team uses.
- It is a conversation layer, not an automation layer. Standard HubSpot API rate limits apply underneath, so heavy, scheduled, or high-volume automation still belongs in a purpose-built integration, not a chat window.
So the framing is: the official MCP path is excellent for natural-language reading and standard-record updates, and it is not yet a full automation layer. For custom objects or deeper workflows you are still looking at the API or third-party tooling. That is not a knock. It is just the line you should walk into the decision knowing.
What about security?
Once an AI tool can read your CRM, every tool call returns whatever that user can see, and your CRM holds PII, contracts, and financial detail. The permission-respecting design is the right foundation: an admin has to enable access for the organization, every connection honors the individual user’s permissions, and HubSpot states that CRM data and customer segments cannot be bulk-copied out through the connector. You should still decide deliberately which users get connected, to which AI tools, and review the data-use terms on the AI vendor’s side for your specific plan. For a small SaaS team this is usually a one-meeting decision. The point is to make it a decision rather than a discovery.
What does this unlock for a RevOps team?
The practical magic is that the busywork between a question and an answer largely disappears:
- Ask your pipeline a question in Claude or ChatGPT and get an answer from live data instead of building a report.
- Update a deal, log a meeting note, or create a contact from the same conversation where you drafted the follow-up, so CRM hygiene becomes a side effect of work you were already doing.
- Triage tickets, check campaign-to-deal attribution, or spot stalled deals conversationally, then push the fix back into HubSpot without switching tabs.
It does not replace HubSpot’s own reporting, and it does not replace clean process. It removes friction. That is the right expectation to set.
Where does the AI actually live, hub by hub?
Short answer: Breeze is not one feature you switch on. It is woven through every hub, and the highest-leverage entry point is different in each.
- Sales Hub: the AI Meeting Notetaker and call summaries, rules-based lead scoring on Professional climbing to predictive scoring and conversation intelligence on Enterprise, Smart Deal Progression connecting transcripts to next steps, and the Prospecting Agent doing the account research your SDRs currently do by hand. (We covered the features most teams miss in our Sales Hub underrated features guide.)
- Marketing Hub: AI content generation, Loop Marketing workflows in the Assistant, personalization driven by CRM data, and now HubSpot AEO for tracking your visibility in AI-generated answers. (Related: our Marketing Hub underrated features guide.)
- Service Hub: the Customer Agent is the headline, resolving the majority of tier-one volume across channels, plus a Knowledge Base Agent that finds documentation gaps and drafts help articles to fill them.
- Content Hub: AI content creation, brand voice enforcement, translation, and image generation built directly into the content tools.
- Data Hub (formerly Operations Hub): the Data Agent and AI-assisted data quality recommendations. The least glamorous application and arguably the most important, because it works on the foundation everything else depends on.
- Commerce Hub: AI surfaces inside the quote, invoice, and payment workflows rather than as a standalone agent.
Why does data quality decide whether any of this works?
Short answer: Every HubSpot AI feature, and every external AI you connect through MCP, is only as good as the CRM data underneath it. AI on a messy portal does not surface insight. It surfaces confident nonsense, fast enough that people trust it before they should.
This is the part the demos skip. Predictive scoring needs clean, consistent deal history. The Prospecting Agent needs complete account records to ground its outreach. Smart Deal Progression can only suggest updates against fields that exist and mean something. And an LLM querying your pipeline through MCP will faithfully report whatever mess is actually in there, with perfect grammar and total confidence.
Outcome-based pricing quietly raises the stakes here in a way few people have noticed. You now pay per resolved conversation and per recommended lead, which means bad data does not just produce bad answers. It produces billable bad answers. An agent recommending leads off stale, duplicated records is an expense, not an asset.
This is why, at SwyftRev, we treat AI readiness as a data hygiene question before it is a feature question. The order is not negotiable: get the foundation clean and consistent, then turn the AI on. Teams that do it in that order get leverage. Teams that do it backward get automated mistakes.
Where should you start with HubSpot AI?
A pragmatic sequence for a B2B SaaS team in mid-2026:
- Turn on Breeze Assistant today. It is free on whatever tier you are on, and Slack availability removes the last excuse.
- Check your enrichment. If you are paying a third party for basic firmographics, standard enrichment is now included with your Core Seats. Reallocate that budget.
- Trial one agent against real volume. The 28-day free trials exist for exactly this. Customer Agent if support volume is your pain, Prospecting Agent if top of funnel is.
- Connect your portal to Claude or ChatGPT through the MCP server and feel what it is like to query your pipeline in plain English. Decide the access question deliberately first.
- Before any of it: audit your foundation. If part of you suspects your data is not clean enough to trust what the AI tells you, that suspicion is the most valuable signal in this whole article.
That last question, “Is our HubSpot foundation ready for AI to be trusted on top of it,” is exactly what a SwyftRev Foundation Audit answers. For a fast first read on where you stand, run the SwyftRev ROI Snapshot.
Frequently Asked Questions
Is HubSpot's AI (Breeze) free?
Partially. Breeze Assistant is free on every HubSpot tier, including the free CRM, and standard data enrichment is included with Core Seats on Starter and above. The autonomous Breeze Agents require Professional or Enterprise and charge per outcome: $0.50 per resolved conversation for Customer Agent and $1 per recommended lead for Prospecting Agent as of April 2026.
How much do HubSpot Breeze Agents cost in 2026?
As of April 14, 2026, pricing is outcome-based. Customer Agent costs $0.50 per resolved conversation (50 credits) and Prospecting Agent costs $1 per lead recommended for outreach. Credits cost $10 per 1,000, plans include monthly allowances (500 on Starter, 3,000 on Professional, 5,000 on Enterprise), and both agents offer a 28-day free trial.
What is the HubSpot MCP server?
The HubSpot Remote MCP Server is a HubSpot-hosted Model Context Protocol server, generally available since April 2026, that lets MCP-compatible AI tools like Claude, ChatGPT, Gemini, and Microsoft Copilot read, create, and update standard CRM records in natural language while respecting each user’s existing HubSpot permissions.
Can ChatGPT or Claude access HubSpot custom objects?
No, not yet. HubSpot’s AI Connectors and MCP server support standard objects only: contacts, companies, deals, tickets, line items, products, quotes, invoices, orders, subscriptions, and segments. Custom objects and unstructured data are not currently exposed, so teams with custom data models still need the API or third-party tooling for that data.
Can AI delete records in my HubSpot portal?
No. Delete functionality is deliberately excluded from HubSpot’s AI Connectors and MCP server to protect data integrity. The connectors support read, create, and update operations only, and every action is limited to what the connected user’s HubSpot permissions already allow.
What is the difference between Breeze Assistant and Breeze Agents?
Breeze Assistant is a free conversational helper: you ask, it answers or drafts. Breeze Agents are autonomous workers on Professional and Enterprise tiers that complete multi-step jobs in the background, like resolving support tickets or researching and contacting prospects, and they charge per completed outcome.
What did HubSpot announce at the Spring 2026 Spotlight?
On April 14, 2026, HubSpot announced 99 updates, including HubSpot AEO (a tool for tracking brand visibility in AI-generated answers, now in public beta), Smart Deal Progression, a rebuilt Prospecting Agent, email support for Customer Agent, Breeze Assistant upgrades including Loop Marketing training and role-aware responses, and general availability of its AI Connectors and Remote MCP Server.
Do I need a paid plan to connect HubSpot to Claude or ChatGPT?
Not on the HubSpot side: the MCP server and AI Connectors work on every HubSpot tier, including the free CRM, at no separate HubSpot charge. The one-click connectors generally require a paid subscription on the AI platform side, such as Claude Pro or ChatGPT Plus, so check current requirements for the specific tool your team uses.
Sources and verification
Verified in June 2026 against HubSpot’s own documentation and the Spring 2026 Spotlight release (announced April 14, 2026). AI features and pricing are changing unusually fast in this area; confirm current details on HubSpot’s live pages before acting.
- Understand Breeze (HubSpot Knowledge Base)
- HubSpot MCP server documentation
- HubSpot Spring 2026 Spotlight (HubSpot product announcements)
- HubSpot Product and Services Catalog
Independent analysis from SwyftRev, a HubSpot Solutions Partner and RevOps consultancy. Claude is a product of Anthropic; ChatGPT is a product of OpenAI.
