GTM-engineering deep dive · MCP + HubSpot · 2026

HubSpot + Claude via native MCP — the broadest CRM-category MCP surface in 2026

HubSpot ships a hosted MCP server covering the full GTM platform — CRM, Marketing Hub, Sales Hub, Service Hub, Operations Hub, Content Hub, and the Breeze AI agent suite (Prospecting, Customer, Content, Social, Data). Public Beta as of 2026. For GTM engineers driving daily orchestration from an AI client, this is the broadest LLM-native surface in the CRM category — one MCP, full platform.

This page is the operator-grade walkthrough: setup, five concrete Claude + HubSpot workflows we've validated, the Breeze credit gotcha that catches everyone in week one, the Public Beta caveats, and the multi-MCP orchestration pattern (HubSpot + Apollo + Smartlead in one Claude session) that collapses the cross-tool friction Zapier was built to bridge.

MCP scope
Full platform
CRM + Marketing + Sales + Service + Ops + Content + Breeze
Status
Public Beta
OAuth, hosted by HubSpot
Free CRM access
Yes
Unlimited users, real product (not a trial)
Breeze AI
MCP-exposed
$1/qualified lead outcome pricing

TL;DR

Want to try HubSpot?

HubSpot ships the broadest CRM-category MCP — install during Public Beta, ship as it goes GA

Free CRM real (unlimited users), paid hubs from Starter $15-$20/seat/mo, Breeze AI bundled into Pro+ tiers with outcome-based pricing. The MCP layer covers everything.

Start with HubSpot →Affiliate link — StackSwap earns a commission if you sign up for HubSpot. We only partner with tools we'd recommend anyway.

What makes HubSpot MCP structurally different

Close MCP covers Close-shaped data (CRM + telephony + Chloe). Attio MCP covers Attio-shaped data (relational object model CRM). HubSpot MCP covers the full GTM stack because HubSpot's product is the full GTM stack on one platform. The breadth shows up in cross-hub workflows that no other MCP-shipping CRM can match.

Realistic example: "which paying customers have open Service Hub tickets AND stalled renewal opportunities in Sales Hub AND haven't engaged with Marketing Hub emails in the past 30 days?" That's a synthesis across three hubs in one conversation. With Close or Attio MCP, you'd need to bolt on a marketing automation MCP and a service desk MCP separately, then orchestrate across them. With HubSpot MCP, the data lives on one platform and one MCP exposes all of it.

Breeze AI agents — orchestrate from Claude

Breeze is HubSpot's bundled AI agent suite. Prospecting Agent (autonomous SDR research + outreach drafting), Customer Agent (AI service rep), Content Agent (marketing asset generation), Social Agent (post scheduling + drafts), Data Agent (enrichment + cleanup). Repriced to outcome-based credits in April 2026: $1 per qualified lead for Prospecting Agent (100 credits), similar outcome pricing across the suite.

The MCP layer exposes Breeze agent invocations as first-class tools. Realistic workflow: "have Breeze Prospecting Agent research and draft outreach to these 50 accounts, queue the drafts for human review." Claude invokes Breeze via MCP; Breeze runs inside HubSpot with full HubSpot context (contact records, past engagement, deal stage); the drafts queue back to Claude for editing or to a HubSpot Sales Hub user for human send. The composition — Breeze for the autonomous work inside HubSpot, Claude for the orchestration across the broader stack — is the workflow shape that justifies running both an external LLM and Breeze in 2026.

Five concrete Claude + HubSpot workflows you can ship today

1. Marketing-to-sales handoff summarization

Weekly Monday-morning prompt: "Summarize this week's MQL conversion in HubSpot. Which contacts converted to SQL, where did the funnel leak, which workflows underperformed against the 30-day baseline?" Claude reads Marketing Hub + Sales Hub in one query, returns a structured summary, and (with a follow-up) suggests where to adjust.

2. Stalled-deal re-engagement drafting

"Find deals in Sales Hub that have been in Demo Scheduled for more than 21 days with no touch. Read the contact lifecycle + last 5 activities + deal notes, draft personalized re-engagement emails grounded in the actual context." Claude reads multiple entities per deal, drafts contextual copy, queues as Sales Hub tasks for human review.

3. Breeze Prospecting Agent orchestration

"Have Breeze Prospecting Agent research and draft outreach to these 50 accounts." Claude invokes Breeze via MCP, monitors completion, reads the drafts, edits where the copy needs sharpening, queues for human send. The agent-on-agent composition.

4. Cross-hub renewal-risk surfacing

"Which paying customers have open Service Hub tickets in the past 30 days AND stalled renewal opportunities AND haven't engaged with Marketing Hub emails?" Three-hub synthesis in one prompt. Surface the at-risk accounts before the CSM team realizes.

5. Content Hub asset generation with brand consistency

"Draft a landing page for our Q2 campaign. Read our existing Content Hub assets to ground in brand voice, generate the copy, create the page in HubSpot as draft." Claude reads existing Content Hub assets via MCP, drafts new copy, creates the page under draft status for human review.

Setup — 1-2 minutes via OAuth

  1. Add HubSpot MCP to your client. Endpoint listed at developers.hubspot.com/mcp. For Claude Desktop or Claude Code, add the server entry. For ChatGPT and Cursor, same flow.
  2. Authorize OAuth scopes conservatively. Start with read scopes only — crm.objects.contacts.read, crm.objects.deals.read, marketing.workflows.read. Expand to write scopes once you have calibration on how the LLM uses the surface.
  3. Connect with a scoped HubSpot user. Not your Super Admin. Create a separate HubSpot user with limited hub access and capped Breeze allocation.
  4. Verify connectivity. "List my recent HubSpot deals." Confirm response, then test a Breeze invocation under explicit confirmation.
  5. Watch the Breeze credit dashboard for the first week. Develop calibration on how the LLM batches agent invocations.

The Breeze credit-burn gotcha

Same shape as Apollo MCP's credit-burn warning. The MCP surface makes it trivially easy for the LLM to invoke Breeze Prospecting Agent against a large account list — each qualified lead = $1 (100 credits). An enthusiastic agent loop can burn through credits faster than expected.

Mitigations: (1) connect with a non-admin HubSpot user with capped Breeze allocation; (2) system-prompt a confirmation gate before any Breeze invocation against more than 25 accounts; (3) watch the Breeze credit dashboard for the first week of MCP use. HubSpot's per-agent credit visibility is sharper than Apollo's enrichment-credit visibility, which helps once you're calibrated.

When HubSpot MCP doesn't unlock value

Same honesty as the other integration guides. If your daily orchestration doesn't run through Claude / Cursor / ChatGPT, the MCP layer isn't adding value to your motion. For non-AI-forward operators, evaluate HubSpot on the structural wins that hold regardless of MCP: the unified contact graph across hubs, Free CRM (real product, unlimited users), and Breeze AI agents bundled into Pro+ tiers.

Want to try HubSpot?

If marketing + sales + service share contact records, HubSpot MCP is the cleanest LLM-native shape

Free CRM real; paid hubs from Starter $15-$20/seat. The MCP covers everything you pay for, including the Breeze AI agents.

Start with HubSpot →Affiliate link — StackSwap earns a commission if you sign up for HubSpot. We only partner with tools we'd recommend anyway.

FAQ

HubSpot MCP is hosted by HubSpot at the platform level. In Claude Desktop or Claude Code, add an MCP server entry pointing at the HubSpot MCP endpoint (current URL listed at developers.hubspot.com/mcp); the MCP client opens a browser tab for the OAuth flow with HubSpot. Authorize the scopes the agent will need (start conservative — crm.objects.contacts.read, crm.objects.deals.read; expand to write scopes once you have calibration). Same OAuth flow works for ChatGPT and Cursor. Total time: 1-2 minutes including the scope selection.

No surcharge for the MCP layer itself. You pay for HubSpot (Free CRM real, paid hubs from Starter $15-$20/seat/mo) and the MCP is included. The cost driver to watch is Breeze AI credit consumption — when Claude invokes Breeze Prospecting Agent via MCP, those credits ($1 per qualified lead, 100 credits) come out of your Breeze allocation. An enthusiastic agent loop can burn through credits faster than expected if you don't set guardrails.

HubSpot MCP is in Public Beta as of 2026. Practical implications: (1) the operation catalog is iterating — workflows that work today may shift; (2) for change-controlled environments (regulated industries, enterprise procurement), treat the Public Beta status as 'evaluate now, deploy at GA'; (3) for SMB and mid-market production deployments, the Public Beta is safe to use with the standard MCP operator setup (scoped user, conservative scope, monitoring). HubSpot has a strong record of moving betas through to GA cleanly; the surface should stabilize through 2026.

Five validated workflows: (1) marketing-to-sales handoff summarization — "summarize this week's MQL conversion, surface where the funnel leaked"; (2) stalled-deal re-engagement drafting — LLM reads contact lifecycle + recent activity + deal context, drafts personalized re-engagement; (3) Breeze agent orchestration — have Breeze Prospecting Agent research and draft outreach to a list of accounts, Claude reviews and edits before send; (4) cross-hub queries — "which paying customers have open Service Hub tickets AND stalled renewal opportunities"; (5) Content Hub asset generation with brand consistency. Workflows we'd avoid: anything that bulk-modifies workflow enrollments without explicit confirmation (high blast radius), anything that fires 100+ Breeze invocations without guardrails (credit burn).

Yes — this is the multi-MCP outbound pattern. Configure Claude with HubSpot MCP, Apollo MCP (native, OAuth, in Claude's connector directory), and Smartlead MCP (community-built by LeadMagic). Claude reads target lists from HubSpot, sources additional contacts from Apollo, pushes them into Smartlead for email cadences, and writes reply detection back to HubSpot as contact + deal activity. The composable shape eliminates the Zapier middleware bill and the per-tool tab-flipping. Pair with Close MCP if you're running a phone-led layer alongside HubSpot.

Order-of-magnitude friction difference. With Zapier as middleware: AI client → webhook → Zapier → HubSpot REST API → response back. Each hop adds latency, each Zapier seat is a monthly line item, and the HubSpot integration in Zapier is maintained by Zapier so schema changes propagate on Zapier's timeline. With HubSpot MCP: AI client → HubSpot MCP → response. One hop. Sub-second. HubSpot maintains its own MCP server so schema updates ship coordinated. Same logic applies for Make, n8n, Workato.

Limited. If your daily orchestration doesn't run through Claude / Cursor / ChatGPT and your team works the HubSpot UI directly, the MCP layer isn't adding value to your motion. You're paying for capability you won't use — though the MCP layer is included on every HubSpot tier so the 'don't pay for it' framing is moot. For non-AI-forward operators, evaluate HubSpot on the structural wins that hold regardless of MCP: the unified contact graph across Marketing + Sales + Service, the Free CRM (real product, unlimited users), and the Breeze AI agents bundled into Pro+ tiers with outcome-based pricing.

Officially supported by HubSpot. Endpoint hosted by HubSpot, OAuth flow handled by HubSpot, docs at developers.hubspot.com/mcp. Separate from this is the local Developer MCP HubSpot ships for CLI and developer-environment work — that's a different server. Both are first-party HubSpot, but they serve different use cases: the hosted MCP for AI-client workflows against HubSpot data, the Developer MCP for building custom HubSpot apps.

Related reading

Canonical URL: https://stackswap.ai/hubspot-mcp-claude-integration. Disclosure: StackSwap is a HubSpot affiliate.