- Yes, HubSpot has MCP servers, and there are two of them. A remote server for connecting AI tools to CRM data at mcp.hubspot.com, and a local CLI-based Developer MCP server for building on the HubSpot Developer Platform.
- The access model is the part that catches people out. The new Developer Platform is required, an admin has to connect the account first, and a user-level app with the right read scopes has to exist before any agent sees anything.
- Sensitive Data Properties are off limits, and association access is bounded to the same documented object set.
- The open question is cost at scale, not capability. Conduyt exposes a 610+ endpoint public API catalog and an included MCP server on flat pricing, so the number of people using AI tooling does not change what you pay.
By Jordan Tate, Head of Growth at Conduyt. Updated August 2026.
Buyers keep asking us this one, and a lot of the answers online are out of date. HubSpot does have MCP servers, they are documented, and they work. So the honest version of this page is not a takedown. It is a straight account of what HubSpot ships, what it does not reach, what you have to do to turn it on, and the one dimension where the comparison with a flat-rate CRM actually bites. Everything about HubSpot below is sourced to HubSpot’s own MCP documentation.
Does HubSpot have an MCP server?
Yes. HubSpot publishes two distinct MCP servers. The HubSpot MCP Server (Remote) gives MCP-compatible AI tools secure read and write access to CRM data through the mcp.hubspot.com endpoint, and the Developer MCP Server (local) is a CLI-based server that lets agentic development tools interact with the HubSpot Developer Platform. Both require the new HubSpot Developer Platform, and the Developer MCP server specifically requires Developer Platform v2025.2.
The two servers, and which one you actually want
HubSpot MCP Server (Remote). This is the one most people mean. It is a hosted bridge that lets an authorized MCP client connect to a specific HubSpot account and work with live CRM records. If your goal is an AI assistant that can look up a deal, summarise a ticket history, or write a note back to a contact, this is the server.
Developer MCP Server (local). CLI-based and aimed at developers building on HubSpot. It helps agentic development tools scaffold and manage projects on the Developer Platform. Useful, but it is a build-time tool, not a way to put CRM data in front of a revenue team.
Confusing the two is the most common mistake we see in buyer conversations, usually in the direction of someone believing they have live CRM access when what they set up was the local developer tooling.
What HubSpot data the MCP server can read and write
What HubSpot data can the MCP server actually read and write?
HubSpot documents read and write access to CRM objects including contacts, companies, deals, tickets, carts, products, orders, line items, invoices, quotes, subscriptions and segments, plus engagements: calls, emails, meetings, notes and tasks. A second tier is read-only: organizational context such as users, teams, reporting structures, owners, roles and seats, and marketing content such as campaigns and campaign metrics, landing pages, website pages and blog posts. HubSpot states the accessible set will grow as more MCP tools are added.
That is a genuinely broad surface and it deserves credit. Read and write on the full core object set plus engagements covers most of what a sales or service workflow needs. The read-only tier is a sensible line to draw: an agent can see who owns what and how a campaign performed without being able to restructure your team or republish your website.
The limits, stated plainly
What can the HubSpot MCP server not reach?
HubSpot states the MCP server does not allow access to custom Sensitive Data Properties, including Personal Health Information and other highly sensitive data. Association access is limited to the same object set that is otherwise exposed. Anything outside the documented object list is not reachable through the server today, which matters if your workflow depends on custom objects or properties that sit in the restricted category.
This is a reasonable restriction rather than a flaw, and any CRM handling regulated data should have one. It is worth knowing before you design a workflow around a property that turns out to be invisible to the agent.
Setting it up, and the parts that stall projects
How do you set up and control access to the HubSpot MCP server?
For the remote server you must be on the new HubSpot Developer Platform, then create a user-level application with read scopes for the CRM objects you want to reach, and use those OAuth credentials to connect to the mcp.hubspot.com endpoint from any compatible MCP client. The account admin has to connect first before other users in the account can. HubSpot supports OAuth 2.0 today and has said clients will need OAuth 2.1 with PKCE and refresh token rotation, so plan for refresh tokens that change on each renewal. The Developer MCP server runs locally, authenticates through the CLI, and is set up with the hs mcp setup command.
Three things in there predict most of the delay. The Developer Platform requirement means an account on the older platform has a migration in front of it. The admin-connects-first rule means a single person becomes the bottleneck for the whole account. And the OAuth 2.1 direction means an integration built today against 2.0 has maintenance coming.
When you are ready to move, the migration playbook is the step-by-step for leaving a legacy CRM without losing a record.
MCP and an API are not the same thing
Is an MCP server the same as an API?
No, and the difference decides what your AI agent can do unaided. An API is a set of endpoints a developer writes code against. An MCP server exposes capabilities in a form an AI client can discover and call on its own, with scoped permissions, so the agent can take an action without someone building a bespoke integration first. A CRM can have an excellent API and no MCP server, in which case connecting an agent means custom middleware. HubSpot has both.
This is why the answer for eLead is different from the answer here. A restricted API with no MCP layer leaves you building middleware. HubSpot does not have that problem.
Where the comparison actually bites
How does Conduyt’s MCP support compare with HubSpot’s?
Both expose CRM data to AI clients through MCP, so the useful comparison is not whether the capability exists but what it costs to use at scale and what surface it covers. Conduyt’s public API catalog carries 610+ endpoints across its domains and its MCP server is included rather than gated behind a platform tier, and Conduyt prices flat rather than per seat, so adding people who use AI tooling does not change the bill. HubSpot’s MCP servers require the new Developer Platform and inherit HubSpot’s seat-based pricing model. Which fits better depends on your seat count and whether you already run on HubSpot.
If you are already deep in HubSpot with a small team, its MCP servers are capable and the migration maths will rarely favour moving. The case changes when headcount grows, because seat-based pricing scales with the number of people you want using AI tooling and flat pricing does not. We have written up how HubSpot’s per-seat pricing works and what HubSpot actually costs if you want to run that comparison against your own seat count, and our HubSpot alternatives page covers the wider decision.
Frequently asked questions
Does HubSpot have an MCP server?
Yes. HubSpot publishes two distinct MCP servers. The HubSpot MCP Server (Remote) gives MCP-compatible AI tools secure read and write access to CRM data through the mcp.hubspot.com endpoint, and the Developer MCP Server (local) is a CLI-based server that lets agentic development tools interact with the HubSpot Developer Platform. Both require the new HubSpot Developer Platform, and the Developer MCP server specifically requires Developer Platform v2025.2.
What HubSpot data can the MCP server actually read and write?
HubSpot documents read and write access to CRM objects including contacts, companies, deals, tickets, carts, products, orders, line items, invoices, quotes, subscriptions and segments, plus engagements: calls, emails, meetings, notes and tasks. A second tier is read-only: organizational context such as users, teams, reporting structures, owners, roles and seats, and marketing content such as campaigns and campaign metrics, landing pages, website pages and blog posts. HubSpot states the accessible set will grow as more MCP tools are added.
What can the HubSpot MCP server not reach?
HubSpot states the MCP server does not allow access to custom Sensitive Data Properties, including Personal Health Information and other highly sensitive data. Association access is limited to the same object set that is otherwise exposed. Anything outside the documented object list is not reachable through the server today, which matters if your workflow depends on custom objects or properties that sit in the restricted category.
How do you set up and control access to the HubSpot MCP server?
For the remote server you must be on the new HubSpot Developer Platform, then create a user-level application with read scopes for the CRM objects you want to reach, and use those OAuth credentials to connect to the mcp.hubspot.com endpoint from any compatible MCP client. The account admin has to connect first before other users in the account can. HubSpot supports OAuth 2.0 today and has said clients will need OAuth 2.1 with PKCE and refresh token rotation, so plan for refresh tokens that change on each renewal. The Developer MCP server runs locally, authenticates through the CLI, and is set up with the hs mcp setup command.
Is an MCP server the same as an API?
No, and the difference decides what your AI agent can do unaided. An API is a set of endpoints a developer writes code against. An MCP server exposes capabilities in a form an AI client can discover and call on its own, with scoped permissions, so the agent can take an action without someone building a bespoke integration first. A CRM can have an excellent API and no MCP server, in which case connecting an agent means custom middleware. HubSpot has both.
How does Conduyt’s MCP support compare with HubSpot’s?
Both expose CRM data to AI clients through MCP, so the useful comparison is not whether the capability exists but what it costs to use at scale and what surface it covers. Conduyt’s public API catalog carries 610+ endpoints across its domains and its MCP server is included rather than gated behind a platform tier, and Conduyt prices flat rather than per seat, so adding people who use AI tooling does not change the bill. HubSpot’s MCP servers require the new Developer Platform and inherit HubSpot’s seat-based pricing model. Which fits better depends on your seat count and whether you already run on HubSpot.
HubSpot details on this page are drawn from HubSpot’s published MCP documentation and were checked on 19 August 2026. Vendor capabilities change; verify against the source before making a decision.