Short answer: Claude and HubSpot is HubSpot (your CRM and system of record) with Claude, Anthropic’s AI model, working as a reasoning layer on top. They connect through the official HubSpot connector for Claude, built on the Model Context Protocol (MCP), so Claude can read and work with your live CRM data within each user’s HubSpot permissions.
Most B2B companies don’t have a data problem. They have an interpretation problem.
The pipeline is in HubSpot. The numbers are in the dashboards. But the answer to “why is the quarter slipping, and what do we do on Monday?” still takes a meeting, three exports and somebody’s weekend.
That’s the gap Claude closes. It doesn’t replace HubSpot. It sits on top of it and does the reading, connecting and explaining that leadership teams don’t have time for.
This guide is the overview: what the pairing is, how it connects, what it can and can’t do, what to fix first and where to start. Each section links to a deeper article when you want the detail.
The short version
- HubSpot is the system of record. Claude is the reasoning layer on top of it.
- They connect through the HubSpot connector for Claude, which uses MCP. Claude only sees what the connected user can see in HubSpot.
- Claude can view, create and update a lot of HubSpot data, from deals and tickets to blog posts and marketing emails. It can’t delete anything.
- The value is faster, better decisions. Not more dashboards.
- Data quality, permissions and a clear reporting framework matter more than the model or clever prompts.
- Start with one use case (usually leadership reporting), get the foundations right, then expand.
In this guide: the Claude and HubSpot cluster
Why does this pairing matter now?
Because the bottleneck in most B2B revenue teams has moved from collecting data to understanding it.
HubSpot makes it easy to capture contacts, deals, activity and attribution in one place. Dashboards show what happened. They rarely say why, or what to do next. That still lands on a revenue leader, a RevOps manager or a founder with a spreadsheet.
Two things have changed. First, HubSpot now has an official connector for Claude, so live CRM data reaches an AI model without exports. Second, models like Claude are now genuinely good at the work between data and decisions: comparing periods, spotting patterns across objects, weighing risk and writing a clear summary.
None of this needs a new CRM. It works on top of the portal you already run.
What do “Claude”, “HubSpot” and “MCP” actually mean?
Claude is the analyst, HubSpot is the database, and MCP is the secure line between them.
HubSpot: the system of record
HubSpot holds the commercial truth of the business: contacts, companies, deals, tickets, activities, campaigns and attribution. If it isn’t in HubSpot, Claude can’t reason about it through this connection.
Claude: the reasoning layer
Claude is a large language model built by Anthropic. On its own it knows nothing about your business. Connected to HubSpot, it can read your records, compare them, explain what changed and draft the summary a leader needs.
MCP: the connection
The Model Context Protocol is an open standard for connecting AI models to other systems. HubSpot runs a remote MCP server, and the HubSpot connector for Claude uses it. Instead of you pasting data into a chat, Claude asks HubSpot for the records it needs, when it needs them, under the connected user’s permissions.
How does Claude connect to HubSpot?
Through the HubSpot connector for Claude, added from Claude’s connector settings and approved in HubSpot.
According to HubSpot’s documentation (last updated 17 September 2026), the connector:
- Works with all HubSpot products and plans.
- Needs a paid Claude plan: Pro, Max, Team or Enterprise.
- Needs an active HubSpot user. Super Admins and users with App Marketplace Access can connect straight away. Everyone else needs a Super Admin to approve the connector first.
- Connects one HubSpot account per Claude account.
On Claude Team and Enterprise plans, an Owner or Primary Owner has to enable the connector for the organisation first. Each person then signs in individually.
The connection takes minutes. The work that decides whether it’s useful takes longer: clean data, clear permissions and a reporting framework. Our step-by-step guide to connecting Claude to HubSpot using MCP covers the set-up in detail.
What can Claude see and change in HubSpot?
Claude can view, create and update a wide range of HubSpot data, within the connected user’s permissions. It can’t delete records.
HubSpot expanded the connector in September 2026, so it now reaches well beyond the CRM.
| Claude can… | Examples | Notes |
|---|---|---|
| View | Contacts, companies, deals, tickets, activities, plus marketing email, content and campaign data | Limited to what the connected user can see in HubSpot |
| Create and update | Contacts, companies, deals, tickets, leads, custom objects, line items, products and activities (calls, meetings, notes, tasks, emails), plus blog posts, website and landing pages, and marketing emails | Up to 10 records at a time. Every change shows in HubSpot’s audit log |
| Create but not update | Properties, campaigns and marketing events | |
| View only | Invoices, orders, payments, subscriptions and conversations | Read-only through the connector |
| Not do | Delete anything | Not supported |
Two settings matter more than most teams expect.
First, if HubSpot’s Sensitive Data setting is on, Claude can’t access engagement data at all. Second, because the connector respects HubSpot user permissions, a sales rep and a CEO get different answers to the same question if their access differs. That’s the right behaviour. It also means your permission model needs to be right before you roll Claude out.
And here’s the bit worth catching: Claude only stops to ask before it writes if your HubSpot write tools are set to “Needs Approval”. Set them to “Always allow” and changes go straight in. We cover approvals and governance in is the HubSpot connector for Claude secure?
What can Claude actually do inside HubSpot?
It turns CRM data into decisions. The strongest use cases are about interpretation, not content.
The full list, with example prompts, is in what Claude can actually do inside HubSpot. Here are the headline areas.
Reporting and revenue intelligence
This is where most companies should start. Claude reads pipeline, deal and campaign data and answers four leadership questions every week: are we on target, what’s at risk, what’s changed and what needs attention.
It goes further than native dashboards. Forecast confidence, pipeline risk scoring, win and loss patterns and cohort analysis are in advanced revenue reporting you can’t do natively in HubSpot. For why standard reporting stalls at board level, read why most HubSpot reporting breaks down at executive level.
Letting leaders ask the CRM directly
A CEO shouldn’t have to request a report to find out which deals are slipping. With Claude connected, a leader asks in plain English and gets a ranked answer with reasons. The approach is in how CEOs can ask questions directly to their CRM, and the weekly question set is in 15 questions every revenue leader should ask HubSpot every week.
Pipeline, deal and forecast risk
Claude looks across activity, stage age, close dates and engagement to flag deals that have gone quiet or won’t land this quarter. You get a short list with evidence, not a wall of red cells.
Sales execution
Meeting prep, account summaries and follow-up drafts all get better when Claude can see the deal history. Anything going to a customer should still pass through a person first.
Marketing content with CRM context
Generic AI content reads like everyone else’s. Content built from real objections, deal notes and customer language doesn’t. See using Claude and HubSpot to create better marketing content.
Finding what’s broken
Claude can read every record in a portal, not a sample. That makes it good at spotting expired close dates, deals with no contacts and catch-all lead sources. Our guide on how to audit HubSpot for revenue leaks explains what to look for.
Building it into a system
When these use cases run on a schedule, with a prompt library and a fixed reporting framework, you’ve got an AI revenue analyst. We explain the architecture in how we built an AI revenue analyst using Claude and HubSpot.
What can Claude not do?
Claude can’t fix bad data, set strategy or own a customer relationship.
- It can’t clean your CRM by itself. It finds the problems. People still need to agree definitions and fix the process that caused them.
- It can’t replace commercial judgement. It weighs evidence. It doesn’t know your biggest client’s CFO just resigned unless someone logged it.
- It can’t see what HubSpot can’t see. Unlogged calls and offline conversations are invisible to it.
- It doesn’t search by meaning. HubSpot’s MCP server uses CRM search, which doesn’t include vector search. Good property structure still matters.
- It can be wrong. Check the outputs, especially numbers heading to a board.
Claude or HubSpot Breeze?
Usually both. Breeze is HubSpot’s native AI for work inside HubSpot. Claude is stronger for analysis that spans objects and systems.
| Need | Better fit | Why |
|---|---|---|
| Quick in-app tasks (drafting, summarising a record) | Breeze | Built into HubSpot, no set-up |
| Cross-object reasoning (deals, activity and campaigns together) | Claude | Built for open-ended analysis |
| Executive and board reporting | Claude | Explains why, not only what |
| Work across HubSpot and other systems | Claude | Connects to accounting, documents and more |
| Fast adoption by non-technical users | Breeze | Less prompting needed |
| Repeatable, governed analysis | Claude with a prompt library | Same questions, same structure, every week |
The full comparison is in Claude vs HubSpot Breeze: which is better?
What does a week actually look like?
Less assembling. More deciding.
Monday, 8am. A leadership briefing lands before the team meeting: revenue against target, what moved in the pipeline last week, the three biggest forecast risks and the suggested actions. Nobody built it over the weekend.
Wednesday. A sales leader asks which deals over a set value have had no activity for three weeks. Claude returns a ranked list with the evidence. The team works the top five.
Friday. Before the forecast call, the team asks for a reasoned confidence view of the committed number, including which deals it depends on.
None of this replaces judgement. It removes the manual work of building the picture, so people spend their time on the decision. The template for that Monday briefing is in how to create a weekly leadership report using HubSpot and Claude.
Not sure where your portal stands? Our AI-Powered HubSpot Audit reads your live portal, read-only, scores it out of 100 across 13 areas and gives you a prioritised fix plan.
Is your HubSpot data ready for AI?
Only if your team already trusts the reports it produces. AI doesn’t repair weak data. It makes the weakness faster and more confident.
Check five things before you connect anything:
- Records. Duplicates merged, required fields enforced, owners assigned.
- Lifecycle stages. Defined, written down and used the same way by sales and marketing.
- Pipelines. Stages that match how buyers buy, with entry and exit rules.
- Source and attribution. No catch-all “offline” source swallowing most of your contacts.
- Close dates and amounts. Kept current, so forecasts mean something.
If two or more of these are shaky, fix them first. The full method, with a readiness scorecard, is in how to prepare your HubSpot data for AI. If you’re not sure, the RevOps health check lists twelve warning signs.
How do permissions, security and governance work?
The connector respects HubSpot user permissions, so governance starts in HubSpot, not in Claude.
That has practical consequences:
- Get HubSpot roles right first. Claude inherits them. Over-permissioned users become over-permissioned AI users.
- Decide who may write. Claude can now change pages and marketing emails as well as CRM records. Decide which teams may do that through Claude, and set the write tools to “Needs Approval” so every change is reviewed.
- Check the Sensitive Data setting. If it’s on, engagement data isn’t available to Claude. Know this before you design a use case that depends on call or email history.
- Understand data use. HubSpot states that Anthropic doesn’t use data shared through the connector to train its models “except in certain instances”, and points to Anthropic’s consumer and commercial terms. Check which terms apply to your Claude plan.
- Be careful with custom connectors. Anthropic advises connecting only to servers you trust and choosing “Allow always” only for tools you’d trust to run unsupervised. On Team and Enterprise plans, a custom connector set up with a shared credential gives everyone who uses it that credential’s access.
- Write it down. A one-page AI usage policy (who can connect, what they may change, what gets reviewed) is enough for most mid-sized teams.
For a role-by-role permission model and the governance checklist, read is the HubSpot connector for Claude secure?
How do you get started, in the right order?
Fix the foundations, design the questions, then connect. Our PACE methodology puts it simply: strategy before channels, demand before systems, systems before AI.
- Fix CRM data quality. Dedupe, enforce required fields, assign owners.
- Sort pipeline and lifecycle stages so the data means something.
- Make attribution and the forecast method usable. It doesn’t need to be perfect. It needs to be agreed.
- Set permissions and a short AI usage policy. Who can read, who can write and what gets reviewed.
- Design the reporting framework. Write down the exact questions leadership wants answered each week.
- Connect Claude through the HubSpot connector and build against that framework, one use case at a time.
The order matters. In our experience, most AI projects that stall skipped steps one to five.
How do you know if you’re ready?
You’re ready when you can say yes to most of these:
- The team broadly trusts HubSpot reports.
- Deal stages are used with discipline.
- Leadership already asks for reporting, and it takes too long to produce.
- You can name the five questions you want answered every week.
- Someone owns revenue operations.
- HubSpot permissions reflect what each role should see.
If several are missing, the right next step is an audit, not an AI build.
Which use case should you build first?
Usually the weekly leadership briefing. It uses data you already have, it has a clear owner and the value shows up within weeks.
Score each candidate on four things: business value, data readiness, ease of adoption and risk if it’s wrong. Reporting scores well on all four. Customer-facing automation scores high on value but badly on risk, so it comes later.
The scoring model, the five use cases we’d build in order and what each costs to do properly are in where to start with AI in HubSpot.
What are the most common mistakes?
- AI before clean data. The answers sound confident and they’re wrong.
- Ad-hoc prompts. Ten people asking ten versions of the same question get ten answers. Build a prompt library.
- More dashboards. You want fewer, better answers.
- Treating Claude as a chatbot. It’s most useful as an analyst inside a defined process.
- Ignoring permissions. Rolling out before HubSpot roles are right exposes data people shouldn’t see.
- Unreviewed writes. Letting AI update records or pages without a human check erodes trust in the CRM fast.
- AI for appearances. If it doesn’t change a decision, it isn’t worth running.
Where does this create commercial value?
In decision speed, forecast confidence and time handed back to senior people.
Leadership teams stop waiting for reports and start asking questions. Reporting effort falls. Risks surface earlier. Forecast conversations start from evidence rather than optimism.
In one published example, an AI operating system built on HubSpot and Claude for a founder-led B2B firm gave back roughly 15 to 20 founder-hours a month, and the bigger gain was in decision quality. The detail is in the Founder Second Brain. For a finance view of the same idea, see what happens when you connect your accounting, CRM and Google Workspace.
How can Imagine Growth help?
We’ve done 100+ HubSpot rollouts. We’re a HubSpot Gold Solutions Partner and a Claude partner. We usually start in one of three places:
- AI-Powered HubSpot Audit. Claude reads your live portal, read-only. You get a Portal Health Score out of 100 across 13 areas and a prioritised fix plan in about two weeks. From £2,500 / $3,500 / €2,900.
- Custom Claude and HubSpot builds. Scheduled leadership reporting, finance dashboards and integrations. Scheduled reporting starts from £1,980 / $2,650 / €2,300 for around three reports.
- AI agent for HubSpot. Claude wired into your HubSpot, live in four weeks, fixed at £9,950 / $13,200 / €11,400. Read and write access is scoped per object and agreed in week one. Nothing goes to a customer without a person approving it.
Prices and what moves them are in our price guide.
Free until 12 January: a Day One session. Thirty minutes where we look at your HubSpot live with Claude, then send you the three fixes worth making first, in writing, within 48 hours. Book a slot.
Key takeaways
- Claude and HubSpot work best as analyst and database: HubSpot holds the truth, Claude explains it.
- The connector needs a paid Claude plan and respects HubSpot user permissions.
- Claude can view, create and update a lot of HubSpot data, including pages and marketing emails, but it can’t delete. Set write tools to “Needs Approval”.
- Data quality and permissions decide whether you can trust the output.
- Start with leadership reporting. Expand once the team trusts the answers.
Frequently asked questions
What is Claude and HubSpot used for? Mainly revenue reporting and decision support. Leaders use it to understand pipeline changes, forecast risk, deal health and campaign contribution without waiting for a manual report. It’s also used for meeting prep, account summaries and content grounded in CRM data.
Do I need a paid Claude plan to connect HubSpot? Yes. HubSpot’s documentation says the connector needs a Claude Pro, Max, Team or Enterprise plan, plus an active HubSpot user. HubSpot itself can be on any plan.
Can Claude change records in HubSpot? Yes, within limits. It can create and update CRM records such as contacts, companies, deals and tickets, and since September 2026 also blog posts, website pages and marketing emails, up to 10 records at a time. It can’t delete anything. We recommend setting write tools to “Needs Approval” so a person reviews every change.
Does Claude respect HubSpot user permissions? Yes. The connector uses the permissions of the HubSpot user who connected it. Nobody can see or change anything through Claude that they couldn’t see or change in HubSpot.
Is our HubSpot data used to train Claude? HubSpot states that Anthropic doesn’t use data shared through the connector to train its models except in certain instances, and refers to Anthropic’s consumer and commercial terms. Check the terms that apply to your plan before rollout.
Is Claude better than HubSpot Breeze? They do different jobs. Breeze is strong for quick tasks inside HubSpot. Claude is stronger for analysis across objects and systems and for executive reporting. Plenty of teams use both.
What do we need in place before starting? Reasonably clean data, defined lifecycle and pipeline stages, agreed attribution, correct HubSpot permissions and a clear list of the questions leadership wants answered.
How long does it take to see value? If the foundations are in place, a first reporting use case can show value within a few weeks. If they aren’t, the first phase is fixing them, and the audit is the faster route.
Does this replace people? No. It removes manual assembly and gives people better evidence. Judgement, relationships and accountability stay with your team.
