HubSpot MCP Server: Privacy-Compliant AI Agent Connection

What is an MCP server for HubSpot?

An MCP server for HubSpot connects AI agents to your HubSpot instance via the open Model Context Protocol. The model then accesses the HubSpot Public API in a standardised way (contacts, deals, tickets, workflows, emails) within clearly defined tool boundaries and permissions.

HubSpot offers a mature, well-documented REST API with OAuth authentication and fine-grained scopes. That makes custom MCP integrations technically straightforward.

Since 13 April 2026, HubSpot's official remote MCP server at https://mcp.hubspot.com has also been generally available (HubSpot changelog). It signs in via OAuth 2.1 with PKCE and needs an MCP auth app in the HubSpot account (Development > MCP Auth Apps). It reads CRM records, marketing and content objects as well as activities, and writes to core CRM objects such as contacts, companies, deals and tickets, always within the existing user permissions. If the setting for sensitive data is active in HubSpot, the server blocks activity objects such as calls, emails, meetings, notes and tasks. A custom MCP server in front of the Public API remains the alternative if you need your own tools, fixed permissions, an audit log, webhooks or a model of your choice.

Where it gets sensitive is data residency. HubSpot operates data centres in the EU and the US. According to HubSpot, the location of new paying customers is determined by the geolocation of the IP address at sign-up; accounts that use only free features end up in the US. Existing paying customers can request a move. If you have to pay strict attention to data residency from a German perspective (e.g. tax advisors, law firms, healthcare), you should go through the setup with us carefully before the MCP server goes live.

MCP server vs. HubSpot Breeze: the key difference

HubSpot promotes its own AI suite Breeze (Breeze Copilot, Breeze Agents, Breeze Intelligence). The obvious question: why an MCP server if HubSpot already offers AI in the product?

Three reasons why German mid-sized companies opt for an MCP server anyway:

• Data residency: HubSpot decides where and with which models Breeze functions are processed; you should check this in HubSpot's privacy and sub-processor documentation before use. Anyone who has to pay strict attention to data residency or falls under § 203 StGB (German Criminal Code, professional secrecy) often wants to choose the model themselves. Your own MCP server on EU hosting, with Claude via AWS Bedrock in the EU Frankfurt region or Azure OpenAI in an EU region, gives you that choice.
• Model independence: With Breeze you are tied to HubSpot's model selection and roadmap. With an MCP server you swap Claude for GPT for Gemini without changing anything in your HubSpot integration: the server stays the same, only the client changes.
• Cross-system agents: Breeze knows only HubSpot. An MCP setup lets an agent access HubSpot, Lexware/DATEV accounting, Personio and your database in parallel, which is the prerequisite for true end-to-end automation ("lead comes in → check creditworthiness → generate quotation → update CRM").

In practice this means: Breeze is good for isolated HubSpot workflows. MCP is the choice as soon as compliance, model freedom or cross-system needs come into play.

HubSpot AI automation: 6 concrete use cases

1. Pipeline updates and forecasts "Which deals are in closing focus this week?" The agent pulls the current deal stage, last activity and probability, grouped by salesperson. Worth its weight in gold before forecast calls.
2. Preparing email sequences Personalised outreach emails: based on HubSpot contact data (industry, last interaction, lead source), the agent builds drafts that only need to be reviewed and sent.
3. Lead scoring and qualification A new lead arrives and the agent checks company data, matches it against your ICP, and proposes a score, owner and next step. Instead of manual lead routing rules.
4. Reporting on demand MRR change, win rate per salesperson, source performance: the agent builds answers from real HubSpot data, not from last week's slides.
5. Ticket classification and routing Incoming service tickets: the agent recognises topic, priority and language, assigns the ticket to the right team and suggests a first reply.
6. CRM data hygiene Detecting duplicates, identifying missing required fields, flagging outdated contacts: the agent does what nobody likes doing but everybody wants done.

How the integration works

Architecture:

AI agent ↔ MCP server (official remote server mcp.hubspot.com or own server in your EU infrastructure) ↔ HubSpot (OAuth 2.1 with PKCE or Public API with OAuth 2.0, REST/JSON)

Concrete steps:

• Choosing the route: For standard CRM tasks the official remote server is often enough. For custom tools, webhooks, sequences logic or a fixed audit log, we build a dedicated server in front of the Public API.
• OAuth app setup: We set up a private or public HubSpot app with clearly delimited scopes (e.g. crm.objects.contacts.read, crm.objects.deals.write).
• Tool definitions: Typical MCP tools: search_contacts, get_deal_pipeline, update_deal_stage, create_engagement, create_email_draft.
• Sequence builder: For email automation we access HubSpot sequences, so the agent can address existing templates instead of mailing freehand.
• Data routing: If a use case needs special data residency, we first check which HubSpot data centre your account sits in and restrict the workflow accordingly.
• Webhook reaction: Optionally the agent reacts to HubSpot events (new form submission, deal stage change), useful for lead routing and onboarding.

GDPR, hosting & security

HubSpot can be used in a GDPR-compliant way if these points are in place:

• Check the HubSpot data centre: New paying accounts are assigned based on location at sign-up, free-only accounts sit in the US; paying accounts can request a move to the EU.
• Define data categories clearly: Which data lands in HubSpot? Which better does not? We make that transparent before the MCP setup.
• Own MCP server in the EU if the MCP layer is to run in Europe; HubSpot itself runs the official remote server.
• DPA with the AI provider (Claude via AWS Bedrock EU, Azure OpenAI in an EU region, etc.), and with HubSpot itself.
• Scope minimum at OAuth level: the MCP server has no blanket access, only explicitly assigned scopes per use case.
• Logging: Tool calls documented, data exports transparent, for GDPR access requests.

What does it cost?

• Clearly scoped entry point: Pipeline reporting or lead routing work well as a first use case.
• Expansion in sprints: Once the MCP server is in place, further tools cost only a fraction.

Frequently asked questions

Which HubSpot API does the MCP server use?

That depends on the route. HubSpot's official remote MCP server (mcp.hubspot.com, generally available since 13 April 2026) signs in via OAuth 2.1 with PKCE. A custom MCP server uses the HubSpot Public API (REST/JSON, OAuth 2.0). Scopes are granted minimally per use case.

Can HubSpot be used with MCP in a GDPR-compliant way?

Yes, if the building blocks are in place: a checked HubSpot data centre, a DPA with HubSpot and the AI provider, clear data categories and a scope minimum at OAuth level.

Can the agent send emails?

Yes, usually as drafts in existing HubSpot sequences. Direct sending without human approval is possible but is recommended only for non-critical templates.

What data residency does HubSpot have?

HubSpot operates data centres in the EU and the US. New paying accounts are assigned based on geolocation at sign-up, free-only accounts sit in the US. We check this in the audit and align the MCP setup accordingly.

Does this also work with custom objects in HubSpot?

Yes. Custom objects are addressed through the same API mechanisms, and MCP tools can map them cleanly.

How quickly is a HubSpot MCP integration productive?

We set the timeline after the AI Check. HubSpot is easy to access on the API side, and with the official remote server, building custom tools is unnecessary for standard tasks.

Which AI models are supported?

All MCP-compatible ones: Anthropic Claude, OpenAI GPT, Google Gemini, Microsoft Copilot.