What HubSpot only offers through MCP.
HubSpot's own MCP server, passed through by the Data Hub for each business unit. It can do what HubSpot doesn't offer in any public API: analytics across the whole CRM, buying intent signals and visibility in AI search engines.
- Vendor MCP · passed through as is
- Coverage: complete
- 31 tools (per vendor) · 11 of them make changes
Read and change.
Your agent works with exactly the permissions of the connected account. Read tools are quick to enable. Tools that change data stay off until you explicitly turn them on.
What your agent can read
- CRM analytics with sums, groupings, medians and time series, asked as a plain-language question.
- Saved reports.
- Buying intent signals.
- Visibility in AI search engines (AEO), with prompts and recommendations.
- Campaign attribution down to individual assets; content analytics, including who viewed which page.
- Segment previews, data model, properties, owners and records.
What your agent can change
- Create and update records and associations; manage segments.
- Create campaigns and assign assets.
- Draft and publish landing pages, website pages and blog posts.
- Manage AEO prompts.
Only once you enable it, per agent and per tool. If you want, a person also approves each action before it runs.
How teams use HubSpot MCP.
Marketing, sales and RevOps run these as recurring routines or on request.
- 01
Pipeline questions in plain language
“What was the median deal size by source last quarter?”- Run the question as an analysis across the CRM
- Return the result grouped, with a trend over time
ResultAn answer with numbers, without anyone creating a report in HubSpot.
- 02
Buying intent → outreach
Every morning- Pull accounts with new buying intent signals
- Match them against owner and deal status
- Build a list for sales or outbound
ResultSales reaches out to companies while the buying intent is fresh.
- 03
Visibility in AI search
Monthly- Pull AEO scores and recommendations
- Add prompts where your company isn't mentioned yet
- Derive content ideas
ResultA plan for getting your company mentioned in AI answers.
- 04
Which assets drive revenue?
At the end of the quarter- Analyze campaign attribution at the asset level
- Compare the strongest and weakest assets
ResultYou see which assets contribute to revenue according to attribution.
Built-in safeguards
- Separate connection through HubSpot sign-in, with the permissions of the user who signs in.
- In the Data Hub, any tool HubSpot doesn't explicitly mark as read-only counts as one that makes changes. None of them are enabled automatically.
- If you want, changes only run after a person approves them, for example the team lead: in the Data Hub, in Slack or via a link. On top of that, each agent has daily and hourly limits.
- Guardrails for every agent: deletes and changes to more than 10 records need approval by default, an emergency brake kicks in on a run of deletes, and you can set your own rules per tool.
- Each connection has an access level: Read, Write or Admin. Where the vendor allows it, the Data Hub only requests the permissions for that level.
- Credentials are stored encrypted in the Data Hub, and the agent never sees them. Every call is logged.
What it can't do
- Which tools show up depends on your HubSpot plan and the signed-in user's permissions.
- If “Sensitive Data” is turned on in HubSpot, HubSpot's MCP blocks activities and conversations.
- HubSpot only exposes some tools (website pages, blog) to certain AI clients. Reports can be read but not created.
HubSpot decides which tools exist through its own MCP server. The hub passes on everything that server offers and refreshes the list every 24 hours.
Connect HubSpot MCP to your agents.
In 30 minutes, we'll show you the Data Hub with your own tools.