The HubSpot Connector for Claude: Setup, Use Cases, and the Limits Nobody Mentions

Today you're going to learn exactly how to connect Claude to your HubSpot portal — and what to do with it once you have.

I've rolled out the HubSpot connector for Claude across client portals since it launched, and I'll be straight with you: the setup is the easy part. Two minutes, no code, free on every HubSpot tier.

The hard part is everything the setup guides don't tell you. Which of the four connection paths actually fits your team. What Claude can write to your CRM (and what it bypasses when it does). And what happens to your data under GDPR the moment it leaves HubSpot's EU data center.

That's what this guide covers. Plus 25+ prompts you can copy straight into Claude today.

Contents

• What is the HubSpot connector for Claude?
• Which connection path should you pick?
• How to set up the HubSpot connector for Claude (step by step)
• What Claude can read and write in HubSpot
• 25+ prompts you can steal (organized by role)
• The advanced paths: MCP servers and Claude Code
• Claude vs. Breeze vs. ChatGPT: which one should you use?
• Is the connector GDPR-compliant? What EU teams need to know
• The limit nobody puts in their launch post
• My rollout playbook
• Frequently asked questions

What is the HubSpot connector for Claude?

The HubSpot connector for Claude is a native integration that gives Claude a governed line into your HubSpot portal. You ask questions in plain language. Claude searches your CRM, reasons across the records, builds charts, and — since write access shipped — creates and updates records too.

That's the definition. Here's what it feels like in practice.

You type: "Which deals slipped this quarter, and why?" Claude pulls your actual pipeline, compares close dates against stage history, and gives you an answer grounded in your data. Not a generic essay about deal slippage.

Your CRM already knows which deals are stalling, which campaigns made money, and which accounts are drifting toward churn. It just couldn't tell you. Now it can.

A few facts worth knowing before we get practical:

• HubSpot was the first CRM with a native Claude connector (July 2025). It started read-only. Write access arrived in November 2025, and everything went generally available in spring 2026.
• Adoption is past the early-adopter phase. HubSpot reports 20,000+ customers using its Claude and ChatGPT connectors across 23 million CRM records, and partner Vantage Point puts the Claude connector past 100,000 installs. (Vendor numbers — but they match what I see in the ecosystem.)
• Every action respects HubSpot permissions. A rep who can only see their own deals gets answers about their own deals. Nothing more.

One more thing, because people mix this up constantly: the connector is not Breeze, and it's not "HubSpot AI." It's Claude — running in Claude's apps — with a secure line into your portal. Where Breeze fits is covered further down.

Which connection path should you pick?

Short answer: business users want the no-code connector. Engineers want an MCP server. Here's the full decision in one table.

There are four legitimate ways to connect Claude to HubSpot. Most guides bury this choice at the bottom. I'm putting it first, because picking the wrong path is the most common way teams waste their first month.

Not sure? You want the first one. The rest of this guide assumes that, and covers the advanced paths later.

And one prerequisite that surprises almost everyone I walk through this: the connector app costs $0 and works on every HubSpot tier — including free CRM. The paid requirement is on the Anthropic side. You need Claude Pro, Max, Team, or Enterprise. Free Claude accounts can't use connectors at all.

How to set up the HubSpot connector for Claude (step by step)

Short answer: Settings → Connectors → HubSpot → Connect, approve the permissions, toggle it on in a chat. About two minutes. The details below are what keep it from breaking later.

Two prerequisites: an active HubSpot user account (any tier), and a paid Claude plan — Pro, Max, Team, or Enterprise.

And one sequencing rule people miss: a Super Admin (or a user with App Marketplace permissions) has to connect first. Until that happens, nobody else in the portal can.

Here's the process:

Step 1: In Claude, click your profile icon → Settings → Connectors.

Step 2: Click Browse connectors.

Step 3: On the Web tab, select HubSpot → click Add to your team. Close the dialog.

Step 4: Back on the connectors page, click Connect next to HubSpot.

Step 5: Select and log into the HubSpot account you want to use.

Step 6: On the "Connecting HubSpot Connector for Claude to HubSpot" page, select the permissions to allow → click Connect app.

Step 7: You're redirected back to the Claude chat window.

Step 8: In any chat, click Search and tools (bottom left) and toggle HubSpot on.

That's it. Test it: "Summarize my open deals by stage, sorted by close date."

Rolling it out to the rest of the team

After the first Super Admin connects, go to: HubSpot → Settings → Integrations → Connected Apps → HubSpot connector for Claude → Give users access → Give Access.

Now, read this part twice, because I've seen it bite: giving users access can't be undone globally. Once you grant it, each user has to uninstall individually. Decide your pilot group before you click.

(If your account was set up before 28 August 2025, users were auto-granted access and the button is greyed out. Nothing you can do about that one.)

The two dates that break setups

Most "the connector is broken" messages I get trace back to one of these:

• Connected before 20 November 2025? Your connection predates write access and engagement data. A Super Admin has to disconnect and reconnect to pick up Edit permissions — and then every user reconnects individually.
• Want to remove permissions later? There's no partial downgrade. You uninstall and reinstall from the Connected Apps page.

Four settings I change on day one

This list comes from HubSpot's own best-practice guidance, and I agree with every line of it:

1. Set Write tools to "Needs Approval" in Claude's connector tool settings — not "Always allow." Claude proposes the change; you approve it.
2. Use the latest Claude models and turn on Extended Thinking for anything analytical.
3. Update one record before any bulk change. Check it landed correctly. Then proceed.
4. Review every proposed write. The connector bypasses your custom validation rules (more on that below) — which means you are the validation rule.

Rolling this out across a sales team and want the permissions, pilot structure, and governance right the first time? Talk to Superwork — connector rollouts are part of the AI-readiness work I do inside HubSpot portals every week.

What Claude can read and write in HubSpot

Short answer: Claude can read almost every standard object, create and update the core CRM objects, and delete nothing. Custom objects aren't supported yet.

Here's the full access surface, per HubSpot's knowledge base:

Now the limits. I'm listing these plainly because this is the part most posts gloss over — and it's the part that determines whether your rollout goes well:

No deletes. Ever. Claude cannot delete a single record through the connector. This is the right design, and it's the first thing I tell nervous admins.

Bulk operations cap at 10 records. For real bulk work — normalizing a property across thousands of records — the connector is the wrong tool. That's MCP territory (covered below).

Writes bypass your custom validation rules. Pipeline-stage rules, association-label requirements — Claude writes straight past all of them. Of everything in this article, this is the line I most need admins to remember.

No custom objects. Standard objects only, for now.

Sensitive data stays out. Custom Sensitive Data properties and PHI are excluded by design. And if Sensitive Data is enabled in your portal, engagement and conversation data is blocked entirely.

One HubSpot account per Claude account, and all usage counts against your HubSpot API limits.

The counterweight to all of this: every create and update lands in HubSpot's Audit Log, attributed to both the user and the connector. Your compliance team can see exactly who did what, when, through which tool.

25+ prompts you can steal (organized by role)

Here's the thing about this connector: it either becomes a daily habit or it dies quietly after week one. The difference is almost never the tool. It's the prompts.

These are the ones I actually use with clients. Copy them as-is, then adapt.

RevOps and admins (start here)

Nobody says this loudly enough: the best first use of this connector isn't sales assistance. It's auditing.

The data-quality audit:

"Audit all company records in HubSpot: flag records missing Industry, Company Size, or Country; identify duplicate company names (fuzzy match); find contacts not associated with a company; flag deals with no associated contact. Return a summary and a CSV of records needing action. Read-only audit only."

See that last sentence? Ending audit prompts with "read-only audit only" is a cheap, effective guardrail — even with write tools set to Needs Approval. I add it to every audit prompt, every time.

Pipeline hygiene:

"Audit my deal pipeline and flag stages where more than 20% of deals have been sitting for over 60 days."

Cleanup by rep:

"Give me all the deals of every AE that need cleanup: wrong close date, stage drift, amount mismatch."

Lifecycle analysis:

"Track how much time contacts spent in each lifecycle stage last quarter and identify where they stall."

Deal velocity:

"Compare our average deal velocity (first contact to closed-won) this quarter vs. the previous two, broken down by deal size (<$10k, $10k–$50k, >$50k)."

Pro tip: when an audit surfaces thousands of records needing the same fix, don't fix them in chat — the 10-record cap makes that miserable. Move to Claude Code with the MCP server. HubSpot's engineering team publishes official deal-management and crm-data-quality skills for exactly this, and community skills like TomGranot/hubspot-admin-skills will scan a whole portal and grade findings A–F.

Sales

Deal-at-risk triage — the single most-adopted prompt pattern I see:

"Review my active HubSpot deals. Highlight the top 5 at risk of slipping based on close date, deal size, and stage. For each, suggest one specific action this week."

Monday prioritization:

"Summarize my active deals by name, amount, and stage, sorted by closing date."

Pre-call research:

"Summarize this contact's full HubSpot history — every logged interaction, company details, open deals, and the last email they engaged with — before my call."

Call follow-up:

"Summarize this call in 3 bullets, list action items with owners, and draft a follow-up email under 150 words referencing key points and next steps."

Closed-won analysis:

"Analyse my last 20 closed-won deals. Identify common traits: company size, industry, lead source, deal length. Distil into a refined ICP, then find 15 open companies that match."

Rep coaching:

"Rank sales reps by close rate for the last 90 days; for the bottom two, show which pipeline stage has the highest drop-off."

Run MEDDPICC or SPICED? Paste your framework into the prompt and have Claude score deals against it. Community skills exist here too — yeutterg's sales-calendar generates MEDDPICC-gap questions per meeting.

Marketing

Segmentation — HubSpot's own flagship example, and for good reason:

"Find contacts who opened our last email campaign but didn't click through, and generate a pie chart to support follow-up segmentation."

Nurture logic:

"Analyze the distribution of contacts across lifecycle stages. Identify the largest group not in an active campaign and suggest a nurture sequence."

Campaign ROI:

"Correlate campaign engagement with deal outcomes to show which channels drive revenue, not just leads."

Lead scoring design:

"Using our contact and deal data, design a lead scoring model: demographic and firmographic fit criteria, behavioral signals, negative scoring triggers, and suggested MQL, SQL, and fast-track thresholds. Show which existing properties support each criterion and which are missing."

One expectation to set with your marketing team: Claude drafts the copy, the scoring model, the workflow logic. It does not build workflows inside HubSpot. A human implements what Claude drafts. That's a limitation today — and a sensible one.

Service and CS

Ticket triage:

"Show all open tickets assigned to me, sorted by priority and creation date."

Trend spotting:

"Summarize the most common questions across inbox and help desk channels this month to identify trending issues."

Support consistency:

"Compare resolution approaches and outcomes across our support channels this quarter. Where do similar tickets get different answers, and which channel resolves fastest?"

Health vs. expansion:

"Rank accounts by ticket volume against open deal value, and separate likely churn risk from likely expansion signal."

And the small one that quietly saves hours every week: "Create a task to follow up with [account] next Tuesday about contract renewal."

HubSpot's own prompt library has 30+ more. But the prompts are the easy part — knowing which audits your portal needs is the hard part. Want a second pair of eyes on where your data quality really stands? Book a working session with Superwork and I'll run the first audit with you, on your live portal.

The advanced paths: MCP servers and Claude Code

The chat connector covers ninety percent of the week. This section is the other ten percent: automation, bulk operations, and cross-tool builds.

If you're not the technical one on your team, skim this and forward it to whoever is.

The remote MCP server (mcp.hubspot.com)

MCP — Model Context Protocol — is the open standard behind connectors. Think of it as a standardized socket any AI client can plug into. HubSpot's remote MCP server works with any MCP-capable client: Claude, ChatGPT, Cursor, whatever ships next.

Setup: in HubSpot, go to Development → MCP Auth Apps → Create MCP auth app (name, description, redirect URL, icon). That gets you a client ID and secret. Auth is OAuth 2.1 with PKCE. Point your client at https://mcp.hubspot.com.

In Claude Code, it's one line:

claude mcp add --transport http hubspot https://mcp.hubspot.com

Then /mcp to authenticate.

Two things worth knowing before you build on it:

Scopes aren't hand-picked. They're set by the tools available at install plus the permissions you grant — and you reinstall to gain new ones.

The search is deterministic. Up to 5 filter groups × 6 filters (AND within a group, OR across groups), 200 results per page, 100 IDs per direct fetch. It's built on HubSpot's CRM Search API — no vector search. For RevOps work, that's a feature: your counts are real counts, not semantic approximations.

The local npm server (@hubspot/mcp-server)

For Claude Desktop and Claude Code users who want to run against a Private App token.

Create a Private App (Settings → Integrations → Private Apps), add scopes — start read-only, like crm.objects.contacts.read — and copy the access token. It's shown exactly once.

Then add to claude_desktop_config.json:

{ "mcpServers": { "hubspot": { "command": "npx", "args": ["-y", "@hubspot/mcp-server"], "env": { "PRIVATE_APP_ACCESS_TOKEN": "<your-token>" } } } }

Fully restart Claude Desktop. A window reload isn't enough — and that's the most common "it's broken" cause I see with this path.

Security basics, since this path skips OAuth: never commit that config to a repo, never paste the token into a shared doc, rotate it if it leaks. And know this is a beta product under HubSpot's Early Adopter Program terms.

The Developer MCP server

For developers building HubSpot apps and CMS assets — templates, modules, serverless functions — via CLI inside Claude Code, Cursor, or VS Code. Authenticated locally through the HubSpot CLI.

Different animal from the CRM-data servers above. If you're a RevOps or marketing reader, you can skip it entirely.

The middleware layer (and why it still matters)

Here's what none of the four paths do: run without you. The connector only works inside a conversation you start. Nothing watches your pipeline overnight. For triggered automation, you still want middleware or the API:

• Zapier / Make + the Claude API. The pattern I see most: new deal created → Claude writes a one-page account brief → brief lands as a HubSpot note.
• Claude for Excel. The Microsoft 365 add-in, for pivoting and charting HubSpot exports. Your enabled Claude skills carry into it.
• Claude in Chrome. Reads HubSpot on-screen with zero API setup. Convenient — but think through the privacy angle before pointing it at client data.
• Third-party data connectors. Supermetrics, Porter Metrics, Coupler.io, Windsor.ai — for blending HubSpot with ad, web, and billing data. One trade-off: asking Claude to total up a Sheets export invites hallucinated arithmetic, where the connector's deterministic filtering doesn't. Aggregate in the tool, reason in Claude.
• The Anthropic API + HubSpot APIs. Fully custom builds, when you've outgrown everything above.

Claude vs. Breeze vs. ChatGPT: which one should you use?

Short answer: all three, for different jobs. Claude for reasoning and visuals in your workday. ChatGPT's Deep Research for long multi-source analysis. Breeze for automation that runs inside HubSpot without anyone prompting it.

Every team asks me this question, and most articles answer it like a boxing match. Wrong frame.

HubSpot ships official connectors for Claude, ChatGPT, Gemini, and Microsoft Copilot — plus its own embedded AI. By HubSpot's own account, more than half its customers already use multiple LLMs. The question isn't which tool wins. It's which tool gets which job.

The budget numbers, because someone in your leadership team will ask:

• Claude: Pro at $20/month ($17 annual), Max at $100 or $200/month, Team at $25/seat (Standard) or $125/seat (Premium), Enterprise custom.
• Breeze: outcome-based since April 2026. The Customer Agent runs $0.50 per resolved conversation (down from $1.00 per conversation), the Prospecting Agent about $1.00 per recommended lead, the Data Agent about $0.10 per answer. HubSpot's CCO Jon Dick put it in one line: "You pay when it works, full stop." For how those credits stack against seat costs, see our full HubSpot pricing breakdown.

HubSpot reports the Breeze Customer Agent resolves 65% of conversations and cuts resolution time by 39% across 8,000+ activated customers. Vendor numbers — but directionally consistent with what I see in portals.

Two footnotes before you standardize your internal docs: parts of the Breeze lineup have been reported as rebranding under "Agent Hub" in mid-2026 — check what your portal actually shows before you commit to either name (we track that shift in HubSpot's agentic CRM pivot). And yes, ChatGPT's connector actually shipped before Claude's. Neither fact settles which your team should use. Use the stack.

Is the connector GDPR-compliant? What EU teams need to know

Short answer: the connector's permission model, audit logging, and no-training default are solid. The nuance is data residency: HubSpot routes through your EU data center, but Anthropic's processing is largely US-based by default. Document the transfer before the pilot — not after.

I work with Nordic and European B2B teams, so this section is where I'll spend the extra words. Most competing guides wave at "enterprise-grade security" and move on.

First, the parts that hold up well under scrutiny:

• Permissions pass through. Claude sees exactly what the asking user's HubSpot login can see. No escalation, no shared super-view.
• Sensitive data is excluded. Custom Sensitive Data properties and PHI never reach Claude. Turn on Sensitive Data and engagement plus conversation data is blocked wholesale.
• Everything is logged. Creates and updates land in the Audit Log, attributed to user and connector. Super Admins can see who connected, reconnected, and when.
• No training on your data. Anthropic doesn't train on connector data unless a user opts in or submits feedback.
• Admin-gated. Only Super Admins or App Marketplace-permission users can enable it, and write tools can sit behind "Needs Approval."

Now the part almost everyone skips. Routing is not residency.

HubSpot routes connector requests through the data center matching your account. An EU-hosted portal routes via HubSpot's EU data center. Good.

But once your data reaches Anthropic, Anthropic's terms govern where and how it's processed — and default Claude infrastructure is largely US-based. EU CRM residency inside HubSpot does not give you EU-only processing by Claude.

For most B2B use under a standard DPA and SCCs, that's a documented, manageable transfer — the same category of question you already answered when you bought HubSpot (we've covered that ground in our HubSpot GDPR and Schrems II guide). But if your DPIA or your customers require EU-only inference, the honest answer in mid-2026 is: you're looking at Claude through a cloud provider's EU region (AWS Bedrock or Google Vertex) rather than the default Claude apps.

What I tell clients to do, in order: review your Anthropic commercial terms and Trust Center against your existing HubSpot DPA. Document the transfer basis. And decide the EU-only question before the pilot — not after legal finds out about it.

The limit nobody puts in their launch post

One more honest section, because it's the sharpest one in the article.

The connector hands Claude your records. It does not hand Claude your business.

Claude can see 340 open deals worth €4.2M. It cannot see that your definition of pipeline coverage is 3.5× against a segmented target. Or that Q3 is seasonally soft in your market. Or that half of last quarter's closed-lost was one churned reseller.

Ask it for pipeline coverage anyway, and it will improvise a calculation. Confidently. Plausibly. And quite possibly not the way your CFO calculates it.

The fix is context discipline:

• Put your metric definitions, targets, and quirks into the prompt — or better, into a shared Claude Project or skill so the whole team runs the same definitions.
• Ask for record links and citations with every analysis, and spot-check them. Claude can hallucinate; HubSpot's own guidance says as much.
• Treat AI output as the first draft of the analysis. Never the decision.

The teams getting real value from this connector aren't the ones with the cleverest prompts. They're the ones who taught Claude how their business defines things.

While we're on honest limits — the rest of the Claude stack matters here. Profile Preferences and Memory stop you re-briefing Claude every Monday. Projects give your team a shared knowledge base, so connector queries run with context. Skills encode your audit procedures and metric definitions so they activate automatically. And Cowork or Claude Code is where connector insights become deliverables and bulk fixes. Adopt them roughly in that order. Each layer compounds the one before it.

My rollout playbook

This is the checklist I run with clients, in order. Screenshot it.

1. Start read-only. Connect with read permissions and run audits and summaries for two weeks.
2. Set write tools to "Needs Approval" before anyone gets write access. Keep it there.
3. Pilot with one team — RevOps, or one pod of sales — before the global "Give users access" click you can't take back.
4. Test one record before any bulk change. Every time. The 10-record cap limits the blast radius, but validation bypass means bad writes are valid-looking bad writes.
5. Review the Audit Log weekly for the first month. Look for connector-attributed writes nobody remembers approving.
6. Expand write access per object as confidence grows. Deals and tasks first. Engagements later.
7. Keep a human in the loop for stage changes and anything customer-facing.
8. Sandbox-test MCP and API paths before pointing them at production.

Frequently asked questions

How do I connect Claude to HubSpot?

In Claude: Settings → Connectors → Browse connectors → HubSpot → Add to your team → Connect, then log into HubSpot and approve the permissions. A Super Admin (or App Marketplace-permission user) must connect first; other users then get access via HubSpot's Connected Apps settings. Takes about two minutes.

Is the HubSpot connector for Claude free?

The connector app is free on every HubSpot tier, including free CRM. You need a paid Claude plan — Pro ($20/month, $17 annual), Max, Team, or Enterprise. Free Claude accounts can't use connectors.

Can Claude edit or delete my HubSpot data?

It can create and update contacts, companies, deals, tickets, line items, products, and engagements (write is in beta). It cannot delete anything. Bulk changes cap at 10 records, and writes bypass custom validation rules — so set write tools to "Needs Approval."

What HubSpot objects can Claude access?

Read access spans contacts, companies, deals, tickets, line items, products, engagements, quotes, campaigns, landing pages, invoices, payments, subscriptions, lists, website pages, blog posts, users, teams, and conversations. Custom objects aren't supported yet.

Does Anthropic train on my CRM data?

No — not unless a user opts in or submits feedback. Connector data isn't used for model training by default.

Is the connector safe for EU teams under GDPR?

It respects HubSpot permissions, excludes sensitive data, and logs every action. HubSpot routes requests through your account's data center, but Anthropic's processing is governed by Anthropic's terms and is largely US-based by default. Document the transfer under your DPA and SCCs; EU-only inference requires Claude via an EU cloud region (Bedrock or Vertex) instead of the standard apps.

What's the difference between the connector and the HubSpot MCP server?

The connector is the no-code path inside Claude's apps — right for most business users. The remote MCP server (mcp.hubspot.com) is an OAuth surface any MCP client can use, with broader tools, built for automation and cross-tool work. Same data, different altitude.

Should I use Claude, ChatGPT, or Breeze with HubSpot?

All three, for different jobs: Claude for conversational reasoning and visuals, ChatGPT's Deep Research for long multi-variable analysis, Breeze for native automation running on HubSpot triggers without a human prompting it.

Can Claude build workflows or campaigns in HubSpot?

It can create campaigns, landing pages, and marketing events, and draft workflow logic and email copy — but it doesn't assemble workflows inside HubSpot. A human implements what Claude drafts.

The bottom line

Two minutes of setup buys your team a CRM it can actually talk to.

That part is easy, free on every HubSpot tier, and better-governed than most admins expect: permissions pass through, nothing gets deleted, everything lands in the Audit Log.

The value shows up when you treat it as infrastructure instead of a toy. Write access behind approval. Audits before automations. Metric definitions taught once and shared. And the MCP paths reserved for the bulk work the chat connector was never built for.

That's the difference between a team that tried the HubSpot connector for Claude for a week and a team whose Monday morning runs on it.

If you want the second outcome — permissions mapped, pilot scoped, the first data-quality audit run on your live portal, and a governance setup your CFO will sign off on — get in touch with Superwork. I do this with Nordic and European B2B teams every week, and I'll walk through it with your portal open.