On this page
The short version
HubSpot MCP is HubSpot's Model Context Protocol access, a remote MCP server plus official connectors for Claude and ChatGPT that let an AI assistant read and update CRM records within each user's HubSpot permissions. RevSure, the agentic GTM platform, runs Digital Workers on a context layer that joins HubSpot to the rest of the GTM stack, with our own GTM engineers managing each worker day to day.
In October, a marketing operations lead at a test and measurement hardware company described his HubSpot portal on a call with us. The company sells to departments inside very large enterprises, and one aerospace and defense customer alone sits under several web domains, "and so now we have six company records," he said. His sales development team bulk uploads prospect lists into the CRM, which held "689,000 contacts" at the time, and the duplicates had reached the point "where I wish that they just did it outside of HubSpot."
He was already running his own reporting and pipeline coverage in Claude. Connect that portal to Claude through HubSpot's MCP server and the assistant reads exactly what is stored: six companies where the business sees one customer, and contacts that may be the same person under two email addresses. A person reading the answer catches some of that. An agent running the same read every Monday does not.
What is HubSpot MCP?
HubSpot MCP is the set of ways HubSpot exposes CRM and marketing data to AI applications through the Model Context Protocol, the open standard our guide to MCP explains in plain English. HubSpot announced on April 13, 2026, that "the remote HubSpot MCP server is graduating from beta and is now generally available to all HubSpot accounts," according to its developer changelog.
There are three ways in:
- The HubSpot connector for Claude, added from Claude's connector settings.
- The HubSpot connector for ChatGPT, added from ChatGPT's app settings.
- The remote HubSpot MCP server at mcp.hubspot.com, which HubSpot's developer documentation says works "with any MCP client that supports OAuth authentication with PKCE."
A separate developer MCP server, for building HubSpot apps locally, is outside the scope of this guide.
How do you connect HubSpot to Claude?
Per HubSpot's knowledge base article for Claude, the connector needs "a paid Anthropic subscription (Pro, Max, Team, or Enterprise)" and is available on all HubSpot products and plans. In Claude, open Settings, then Connectors, choose Browse, select HubSpot and click Connect. Sign in to HubSpot, choose the permissions and click Connect app.
Super Admins and users with App Marketplace Access can connect straight away. "For all other users to connect, a Super Admin must first approve the connector." HubSpot also notes: "You can only connect one HubSpot account at a time to a single Claude account." Our guide to Claude connectors covers how directory connectors like this one differ from custom ones.
How do you connect HubSpot to ChatGPT?
HubSpot's ChatGPT article says the connector "is available to all HubSpot customers across all tiers with an OpenAI account." In ChatGPT, open Settings, then Apps, select HubSpot and click Connect, then sign in to HubSpot, choose permissions and click Connect app. The same Super Admin approval applies.
HubSpot states that custom validation rules, including pipeline stage validations and association label validations, are not applied through the ChatGPT connector, so a stage change made in chat can skip the checks a rep would hit in HubSpot itself.
HubSpot MCP server and connectors: what each reads, writes and leaves out
Every route respects the user's existing HubSpot permissions. They differ on who can use them, what they write and where they stop.
Two lines in HubSpot's developer documentation explain most of the behavior behind that table: "Behind the scenes, the HubSpot MCP server is based on the CRM search API," which "currently doesn't include vector search capabilities." The server filters and pages through HubSpot records as they are stored. Our comparison of MCP vs API explains why most MCP servers are a layer over an existing API in this way.
What changes when a digital worker reads HubSpot instead of a person?
A person in Claude asks about one account, reads the answer and judges it. A digital worker owns a standing piece of GTM work, such as building account briefs for every target account each week or checking open deals for missing buying group contacts, and it reads HubSpot on every run. Four things shift when the reader is a worker.
Duplicates turn into wrong answers at volume. The hardware company's six company records are a nuisance in one chat, because the person asking knows they are one customer. To a worker ranking accounts by engagement, they are six accounts, each with a slice of the activity, and none of them looks as engaged as the real customer is. That error repeats on every run until something upstream resolves the records.
Associations are rebuilt by the model, call by call. Buying groups and deal influence depend on how contacts, companies and deals connect. HubSpot's query_crm_data tool "does not support joins, column aliases, subqueries, association traversal, or currency conversion," and search returns up to 200 results per page. A worker covering hundreds of accounts makes many calls and stitches the associations together inside its own context window, which is where small mistakes compound.
HubSpot holds part of the account. Ad engagement, intent, product usage and call recordings often live in other systems. CRM, MAP, intent, ads, product and conversations each hold only part of the picture. A worker reading only HubSpot reports the part it can see as the whole.
Writes need rules set in advance. In chat, a person confirms each proposed change, ten records at a time. A worker that updates scores or tasks every week needs its approval points decided before it runs, plus a record of every write. Our guide to MCP security for enterprise AI agents covers the controls security teams ask for.
Sensitive Data adds a fifth wrinkle for regulated teams. With it turned on, the MCP server blocks activity and conversation data, and HubSpot notes that "this restriction is specific to the MCP server and doesn't apply to the standard CRM APIs." An integration built on the CRM APIs can still see engagement history that an agent reading through MCP cannot.
How do HubSpot MCP and the RevSure MCP server work together?
We read HubSpot before any agent does. Our HubSpot integration syncs contacts, companies, deals and marketing engagement, resolves them into one identity per person and account, and joins HubSpot data with ad, web and sales data on the context layer. Agents write lead scores, tiers and recommended actions back into HubSpot under approval.
Our MCP server exposes that context layer to Claude and ChatGPT. Horizontal MCP servers expose raw systems and make the LLM reconstruct what records mean, how the GTM motion works and what the data says. We put that domain knowledge and reusable analytical logic in the tool layer. Every agent call carries a scoped, revocable token, with role-based access, field-level masking and an append-only audit log.
The two servers do different jobs, and many teams will run both. HubSpot's server is the direct route for record-level work in HubSpot: look up a contact, log a call, update a deal. The RevSure MCP server is the route for questions and standing work that span HubSpot and the rest of the stack, such as which accounts are engaging across channels or which open deals lack a buying group. Teams that also run Salesforce will find the Salesforce side in our companion guide to Salesforce MCP. Our overview of MCP for GTM shows how several servers sit side by side in one assistant.
Who runs a digital worker on HubSpot data?
Our Digital Workers are human supervised, and the supervision does not fall on the customer's team. Your team sets the strategy and approves exceptions. Our own GTM engineers manage the digital worker day to day and are accountable for the completion and quality of the work it delivers. They work as an embedded GTM pod: they operate the worker on the customer's own data and tune it as the work changes. Each worker reads and writes to CRM, MAP, ABM/intent and sales engagement directly.
A customer success software company shows where that line falls. It runs Salesforce and HubSpot and already pipes RevSure data into its own warehouse and BI tool. Its marketing operations lead owns the numbers, and before the campaign team could query them through MCP in Claude, he wanted every answer to match the definitions in the company's validated reports. What a campaign manager means by a request, he told our team, "doesn't always reflect like business rule and data definitions." On that call, our team mapped the company's custom HubSpot UTM fields for campaign, medium and source, and built the event pipeline view he asked for. MCP prompts for the campaign team were planned for once access was approved.
The owner set the definitions and the approval point, and our team did the mapping and the build. Digital Workers run on the same split, with our GTM engineers accountable for the work.
How do you check a HubSpot portal before an agent reads it?
Four checks, run before any agent reads the portal on a schedule:
- Run get_user_details at the start of a session, as HubSpot recommends, to see which objects and tools the connection can reach.
- Check whether Sensitive Data is turned on. If it is, the agent will not see engagement history through MCP.
- Count company records per customer across domains and subsidiaries, and contacts that share a name and company under different emails.
- Decide which writes need a person's approval and who reviews the agent's output each week.
What the hundredth run reads
Connecting HubSpot to Claude or ChatGPT now takes a few clicks and an admin approval. The harder question is what an agent reads on its hundredth run, when nobody is checking each answer. If the portal says six companies, a worker reports six. Whether that becomes one customer, with one buying group and one engagement history, depends on what sits between HubSpot and the agent, and on who is accountable for the work the agent hands back.
FAQs
What is the HubSpot MCP server?
The HubSpot MCP server is HubSpot's remote Model Context Protocol server at mcp.hubspot.com. It lets AI applications search, read, create and update HubSpot CRM records, activities and marketing data through tools, within each user's HubSpot permissions. It became generally available to all HubSpot accounts on April 13, 2026, and requires OAuth 2.1 with PKCE for every connection.
Can I connect HubSpot to Claude and ChatGPT?
Yes. HubSpot publishes official connectors for both. The Claude connector needs a paid Anthropic plan (Pro, Max, Team or Enterprise), and the ChatGPT connector is available to HubSpot customers on any tier with an OpenAI account. Users without Super Admin rights or App Marketplace Access need a Super Admin to approve the connector first.
Which HubSpot plans include MCP access?
HubSpot lists its Claude and ChatGPT connectors as available on all products and plans. For the remote MCP server, HubSpot says tool availability varies by subscription, user permissions and account configuration, so some tools, such as quotes or AEO features, need specific paid hubs. Check with the get_user_details tool which tools a connection can reach.
Can the HubSpot connectors delete records?
No. HubSpot's Claude connector supports view, create and update actions but not deletes, and the ChatGPT connector lists no delete access for any object. Bulk creates or updates through either connector are limited to 10 records at a time, and all usage counts against HubSpot's API usage limits.
How is the RevSure MCP server different from HubSpot's MCP server?
HubSpot's server reads and writes the records stored in HubSpot. The RevSure MCP server exposes our context layer, which joins HubSpot with ad, web, intent and sales data and resolves people and accounts, and its tools return computed answers about stages, handoffs and buying groups. Many teams run both side by side in one assistant.
Is it safe to give an AI agent access to HubSpot through MCP?
HubSpot's MCP access respects each user's existing permissions, and its write tools ask the user to confirm changes. For agents that run on a schedule, teams also need approval rules set in advance, scoped credentials and an audit trail of every read and write, which is how we govern calls to our own MCP server.