HubSpot’s hosted MCP is available at mcp.hubspot.com and exposes the CRM, marketing, and service data your private app is allowed to read or write, depending on the scopes you select. You typically authenticate with a private app access token issued in the HubSpot developer settings, then use any MCP client to run natural-language account research, deal reviews, and follow-up drafting.
https://mcp.hubspot.com
Claude Sonnet 4.5
MCP Playground runs 30+ models on the same workflow: switch anytime, or use Compare mode to run several in parallel and balance quality vs. cost.
HubSpot private app access token (Settings → Integrations → Private Apps) with CRM scopes for the objects the agent will touch. Tokens look like `pat-…` or similar per HubSpot’s current format.
How models use it and what it is built for.
In practice the HubSpot MCP lets models search and filter companies, contacts, and deals, retrieve engagement history, update pipeline fields when you approve writes, and turn structured CRM data into short narratives for sales and success teams. The same private-app and OAuth scope model applies as the HubSpot REST API: the agent cannot read objects you have not granted, which makes it a good match for GTM playbooks and pipeline reviews when paired with a carefully scoped private app in a sandbox or production portal.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Which open deals in the negotiation stage have had no activity in 14 days?
Pull the full timeline for this company and the contacts we met last month.
Draft a short follow-up email to this contact referencing their last ticket and deal stage.
List every MQL from last week’s webinar and their current lifecycle stage.
This is not a single-model product: you get the same MCP connection with 30+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for HubSpot
Claude Sonnet 4.5
Claude Sonnet 4.5 is reliable on multi-record CRM questions and natural-language follow-ups. For tight schema-only lookups, Haiku 4.5 is cheaper; compare in Agent Studio.
Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 30+ models, and start chatting — no install required.
Try HubSpot in Agent StudioCommon questions about connecting, scoping and using it safely.
It is a hosted remote MCP HubSpot operates so that AI clients can work with the same CRM, marketing, and service data you already expose to integrations. You supply credentials from a private app, and the model can filter records, read associations, and help draft or apply updates in line with the scopes you granted.
In HubSpot, open Settings, then Integrations, then Private apps. Create a private app, select the CRM and engagement scopes the agent will need, install it to the portal, and copy the private app access token. Paste that token into the MCP client’s access-token field, not a legacy API key you use elsewhere without scopes.
Treat a private app token like any production credential: do not check it into source control, rotate it if you suspect exposure, and start with a sandbox or read-only portal. Grant write scopes only when you actively want the agent to change deals or create engagements.
A token is issued per private app in a specific HubSpot account. You would connect separate instances or change tokens when switching between portals, just as you would for direct API integrations.
Claude Sonnet 4.5 is a strong default for “which accounts are at risk and why” questions that combine activity history, field values, and free-text notes. You can A/B the same question against Haiku 4.5 in Agent Studio if you are mostly listing rows.