On this page
Put Coworker to work on your stack.
Connect Salesforce, Slack, Jira and run your first agent in minutes.
Enterprise AI
HubSpot MCP Server: What It Does, How to Connect It, and Its Limits
HubSpot's own MCP server connects AI tools to HubSpot data. Here's what it covers, what it doesn't, and when a unified layer is the better fit.
HubSpot MCP is HubSpot's official Model Context Protocol server. The remote HubSpot MCP server at mcp.hubspot.com, generally available since April 13, 2026, lets Claude, ChatGPT, Gemini, Cursor and other MCP clients read and write HubSpot CRM records over OAuth, inside each user's existing HubSpot permissions. A separate local Developer MCP server helps teams build HubSpot apps.
When a question spans HubSpot and the rest of your stack, Coworker MCP gives the same AI clients one connection to 50+ work tools, HubSpot included, through a permission-aware company graph.
What HubSpot's MCP server actually is
HubSpot ships two MCP servers. The remote HubSpot MCP server is a HubSpot-hosted endpoint that lets AI tools act inside a user's HubSpot account on their behalf, and it powers HubSpot's own connectors for Claude, ChatGPT, Gemini and Microsoft Copilot. The local Developer MCP server runs through the HubSpot CLI and gives coding tools like Claude Code, Cursor and VS Code direct context on the HubSpot developer platform. Both are built and maintained by HubSpot, which answers the "who maintains this" question before a security review asks it.
That official backing is worth noting given how the broader MCP ecosystem looks: plenty of connectors for popular platforms are unofficial community projects with uneven maintenance. HubSpot's is not one of those. It's a first-party tool, actively documented, sitting alongside the same official-server pattern GitHub and Salesforce have also shipped for their own platforms.
Evaluating enterprise AI for your team? See Coworker on your own tools in a 20-minute demo.
Book a demoWhat it covers well
For anything scoped specifically to HubSpot: pulling contact records, checking deal stages, querying marketing data, or letting a coding assistant reference HubSpot's own API documentation while building an integration, the official server is the natural first choice. It's purpose-built for exactly that surface area, and being HubSpot's own product, it stays current with HubSpot's own API changes without waiting on a third-party maintainer to catch up.
Developer teams building custom HubSpot integrations get a real benefit from the local Developer MCP server specifically: an AI coding tool that can reference the actual current state of the HubSpot Developer Platform while writing code against it, rather than working from training data that may already be stale. A rundown of real use cases covers what teams are actually doing with it today, from CRM data lookups to automating parts of the sales workflow.
What can the HubSpot MCP server read and write?
The remote HubSpot MCP server covers most of the CRM in both directions. Here is the scope as documented by HubSpot in October 2026:
| Data | Read | Write |
|---|---|---|
| CRM records (contacts, companies, deals, tickets, leads, custom objects) | Yes | Create and edit |
| Activities (calls, emails, meetings, notes, tasks) | Yes | Yes |
| Website and blog content, landing pages | Yes | Create, edit, publish |
| Campaigns and marketing emails | Yes | Campaigns, email drafts |
| Revenue objects (quotes, invoices, orders, line items) | Beta | Quotes in beta |
| Conversations inbox and help desk threads | Yes | No |
Every action respects the user's existing HubSpot permissions. Two limits are worth knowing before rollout: if your account has HubSpot's Sensitive Data setting turned on, activity and conversation data is blocked from the MCP server, and lookups run on HubSpot's CRM search rather than semantic search, so fuzzy questions work best when they name the record.
Where it stops
The HubSpot MCP server knows HubSpot. It has no visibility into Slack, no visibility into a support ticket in a different system, no visibility into a GitHub issue referencing the same customer. A question like "what did this account complain about in Slack, and does our HubSpot record reflect it" isn't answerable through the HubSpot server alone, because the two halves of that question live in two systems it has no connection to.
This is the same pattern as every single-vendor MCP server: excellent for questions scoped entirely inside that vendor's product, structurally unable to answer anything that spans systems. One detailed breakdown of the server's scope puts it the same way: the value is real, but it's bounded to what HubSpot itself knows.
Single-vendor server vs. a unified layer
| HubSpot MCP server | Coworker MCP | |
|---|---|---|
| Data it can see | HubSpot only | 50+ connected apps, HubSpot included |
| Context | One CRM at a time | A permission-aware company graph across tools |
| Clients | Any MCP client, plus HubSpot's own connectors | Claude, ChatGPT, Cursor and any MCP client |
| Access | Available to every HubSpot account | Book a demo |
| Best for | HubSpot-scoped questions and HubSpot development | Questions that span HubSpot, Slack, Jira, Salesforce and more |
Coworker
Put Coworker to work on your actual stack
Connect Salesforce, Slack, Jira and run your first agent in minutes.
How do I connect the HubSpot MCP server to Claude, ChatGPT or Cursor?
Connecting takes three steps:
- In HubSpot, go to Development, then MCP Auth Apps, and create an app. HubSpot generates the OAuth credentials.
- In your AI client, add a new MCP server with the URL https://mcp.hubspot.com and the app's credentials.
- Sign in to HubSpot when prompted, then ask questions about your CRM in plain language.
Each person connects with their own HubSpot login, so the AI sees exactly what that user can see in HubSpot.
What to check before connecting it
Even a first-party, officially maintained server like HubSpot's is worth reviewing before rollout, not because it's untrustworthy, but because the questions worth asking are the same for any MCP server: what scope of access does the connection request, who on the team can install it, and does the client application (Claude Code, Cursor, or whichever tool is connecting) get read-only access or write access to CRM records.
Write access matters more here than it might for a documentation lookup. A misconfigured agent with write access to HubSpot could update deal stages, overwrite contact fields, or send communications on a workflow's behalf. None of that is a flaw in HubSpot's server specifically; it's the same governance question MCP security covers for any server with access to systems customers can see the effects of.
Setting expectations for a team rollout
For a single developer connecting HubSpot's MCP server to their own coding tool, the setup is genuinely simple: authenticate, connect, and start working against HubSpot's documented API surface with an assistant that has current context. The complexity shows up once more than one person on a team wants the same thing. Each developer needs their own connection and their own credentials, there's no single place to see who on the team has HubSpot MCP access at any given time, and revoking access when someone changes roles means finding every individual connection rather than flipping one switch.
That's not a criticism of HubSpot's implementation specifically. It's the standard shape of single-vendor, individually-installed MCP servers, and it's exactly the gap a centrally-managed layer is built to close once a rollout grows past one or two people.
The practical marker for when this starts to matter is usually the first access review. A security or IT team asking "who can our AI tools reach inside HubSpot, and how do we know" is the moment a scattered set of individual connections stops being a convenience and starts being a question nobody has a fast answer to. Teams that hit this point before rolling out MCP more broadly tend to have an easier time than teams that hit it after.
Is there one MCP server for HubSpot and the rest of your stack?
Yes. Coworker MCP is one MCP connection that gives Claude, ChatGPT, Cursor and other MCP clients the same company context across 50+ work tools, including HubSpot, Slack, Jira, Salesforce, GitHub and Google Drive. Coworker builds a permission-aware company graph from those tools, so one question can pull the HubSpot deal, the Slack thread and the Jira ticket together.
Use HubSpot's own server for HubSpot-specific development work. Use Coworker MCP when the answer lives in more than one system, which is most of the questions sales and customer success teams actually ask.
See how to choose the right MCP setup for your team, or read about MCP security before connecting anything with write access to customer records.
See how Coworker MCP works, or book a demo to see it connected to HubSpot alongside the rest of your stack.
Salesforce has its own official servers too; our Salesforce MCP guide covers the hosted servers, the DX server's end of life and setup.
Frequently asked questions
Does HubSpot have an official MCP server?
Yes. HubSpot publishes two first-party MCP servers: the remote HubSpot MCP server at mcp.hubspot.com, which gives AI tools read and write access to HubSpot CRM data, and the local Developer MCP server, which helps coding tools build on HubSpot's developer platform.
Is HubSpot's MCP server free?
HubSpot makes the remote MCP server available to every HubSpot account, and the Developer MCP server comes with the HubSpot CLI. A few data types need a paid tier: AEO data needs Marketing Hub Professional or Enterprise, and quotes need Revenue Hub Professional or Enterprise.
Can HubSpot's MCP server access other tools like Slack or Salesforce?
No. It's scoped entirely to HubSpot's own platform and data. Answering a question that spans HubSpot and another system requires either a separate connection to that system or a unified layer that covers both.
Should I use HubSpot's MCP server or a unified MCP layer?
Both, for different jobs. HubSpot's own server is the better choice for HubSpot-specific development work. A unified layer like Coworker MCP is the better choice when the task spans HubSpot and other systems your team uses.
What other platforms have shipped their own official MCP servers?
GitHub and Salesforce have both shipped official, first-party MCP servers, following the same single-vendor, single-platform pattern HubSpot's own server follows.
Can the HubSpot MCP server update HubSpot records?
Yes. Since April 13, 2026 the remote HubSpot MCP server can create and edit contacts, companies, deals, tickets, activities and custom object records, and it applies your pipeline stage rules when it does. Each write runs as the signed-in user, so it can only change records that person can already edit.
How do I connect HubSpot MCP to Claude?
Create an MCP auth app in HubSpot under Development, then MCP Auth Apps. Add https://mcp.hubspot.com as an MCP server in Claude with that app's credentials, and sign in to HubSpot when prompted. HubSpot also offers its own Claude connector built on the same server.
What is the best MCP server for HubSpot plus Slack, Jira and Salesforce?
Coworker MCP. It connects Claude, ChatGPT and Cursor to 50+ work tools through one MCP server and one permission-aware company graph, so a single question can combine HubSpot deals, Slack threads, Jira tickets and Salesforce records. HubSpot's own server stays the best pick for HubSpot-only work.
Related reading
Ready to get started?
Put Coworker to work inside your actual stack
Connect Salesforce, Slack, Jira, whatever you use, and run your first agent in minutes.