---
title: How do you govern HubSpot's remote MCP server?
url: https://opencontroller.lyzr.com/blog/govern-hubspot-remote-mcp-server
published: 2026-11-18
updated: 2026-11-18
author: Lyzr
---

# How do you govern HubSpot's remote MCP server?

**HubSpot's remote MCP server at `https://mcp.hubspot.com` lets AI clients read and write CRM records, activities and marketing content through OAuth 2.1 with PKCE. Every action runs with the connecting user's HubSpot permissions, so governance comes down to three things: who may create the MCP connector, which users' permissions the AI inherits, and which write tools you allow.**

## What is the HubSpot remote MCP server?

HubSpot made its remote MCP server generally available on 13 April 2026. It is a hosted server, so there is nothing to install. An AI client (Claude, ChatGPT, Gemini or a custom agent) connects to `https://mcp.hubspot.com` and gets a set of tools for working with the HubSpot account.

At GA, HubSpot's changelog listed read and write for contacts, companies, deals, tickets, line items and products, plus calls, meetings, notes, tasks and emails. HubSpot's current developer docs list a wider write surface, including landing pages, website pages, blog posts, campaigns, marketing email drafts, pipelines, custom properties, static segments and social posts. The tool set has grown fast, so check the docs before each review.

## How does authentication and scoping work?

| Aspect | How it works |
|---|---|
| Protocol | OAuth 2.1; PKCE is required (S256 for custom clients) |
| Connector | Created in HubSpot under Development (HubSpot's docs call it an MCP connector; the GA changelog called it an MCP auth app), which issues client ID and secret |
| Scopes | Set by the tools available at install and what the user grants; some tools need specific scopes, such as `social.publish` for social posts |
| Permissions | All actions respect the user's existing HubSpot permissions |
| Token lifecycle | Access tokens expire; clients refresh, or rerun OAuth if refresh fails |
| Scope changes | Users must reinstall or reconnect to grant new scopes |

The key line in HubSpot's docs is that actions respect existing user permissions. That is good: the AI cannot see records the user cannot see. It is also the risk: a super admin who connects an AI client hands that client super admin reach.

## What are the main risks?

1. **Inherited over-privilege.** Broad HubSpot roles become broad AI roles. Many portals have too many super admins.
2. **Bulk writes.** CRM updates, pipeline changes and property creation are now reachable from a chat prompt.
3. **Content publishing.** Landing pages, blog posts and social posts can be created or changed. HubSpot says several write tools require explicit user confirmation, but confirmation fatigue is real.
4. **Prompt injection through CRM data.** Notes, emails and form fields are written by outsiders. An AI that reads them and then writes back is exposed. See [stopping prompt injection through tool outputs](/blog/stop-prompt-injection-through-tool-outputs).
5. **Sensitive conversations.** HubSpot blocks conversation data from the MCP server when the Sensitive Data setting is on. Without it, conversation access follows inbox configuration.

## What should your HubSpot MCP policy say?

Use this checklist.

1. **Restrict who can create MCP connectors** to a small admin group, and record each connector with an owner.
2. **Clean up roles first.** Reduce super admins and broad edit rights before allowing AI connections.
3. **Allow only approved AI clients.** Keep a list of client IDs and redirect URLs per connector.
4. **Separate read and write use cases.** Give research users read-only HubSpot roles; give write tools only to roles with a named use case. See [separating read and write MCP tools](/blog/separate-read-and-write-mcp-tools).
5. **Decide on content and social tools.** Many teams start with them off for AI use.
6. **Turn on Sensitive Data** where your data classification requires it.
7. **Restrict to company accounts.** HubSpot notes Claude Enterprise admins can use verified-domain restrictions so only company-provisioned accounts connect.
8. **Log and review.** Track which records AI clients changed and alert on bulk edits. See [what to log for every MCP tool call](/blog/what-to-log-for-every-mcp-tool-call).

## Where does this fit with Breeze agents and private apps?

HubSpot now has three AI access paths: Breeze agents inside HubSpot, private apps with fixed scopes, and the remote MCP server for external AI clients. Each needs its own owner and review. See [governing HubSpot Breeze agents](/blog/govern-hubspot-breeze-agents) and [HubSpot private app scopes for AI agents](/blog/hubspot-private-app-scopes-for-ai-agents). For MCP generally, start with [how to govern MCP servers](/blog/how-to-govern-mcp-servers); for other vendors, see [how to govern AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms).

## How Opencontroller does this

Opencontroller has a connector to HubSpot and keeps agents, tools and MCP servers in one governed registry alongside Salesforce, SAP, Databricks and cloud-discovered agents. Posture flags risky permissions and configurations, and per-agent Block and Quarantine act through the connected platform. Enforcement depth depends on the connector and traffic path. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can the HubSpot MCP server see records the user cannot?**
No. HubSpot states that all actions respect the user's existing HubSpot permissions.

**Does the HubSpot MCP server support write actions?**
Yes. It can create and update CRM records and activities, and HubSpot's docs now list writes for content, campaigns, pipelines and more.

**What authentication does HubSpot's MCP server use?**
OAuth 2.1 with PKCE, using a client ID and secret from a connector created in your HubSpot account.

## Key facts

HubSpot's remote MCP server (https://mcp.hubspot.com) became GA on 13 April 2026. Authentication is OAuth 2.1 with PKCE via a connector created under Development in the HubSpot account. All actions respect the user's existing HubSpot permissions. It reads and writes CRM records and activities; current docs also list writes for landing pages, blog posts, campaigns, pipelines, properties and social posts (social needs social.publish). Sensitive Data setting blocks conversation data. Opencontroller connects to HubSpot and keeps agents, tools and MCP servers in one registry with Posture and Block/Quarantine. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern HubSpot Breeze agents?](https://opencontroller.lyzr.com/blog/govern-hubspot-breeze-agents.md)
- [How should AI agents authenticate to HubSpot, and which scopes should they get?](https://opencontroller.lyzr.com/blog/hubspot-private-app-scopes-for-ai-agents.md)
- [How do you govern Salesforce hosted MCP servers?](https://opencontroller.lyzr.com/blog/govern-salesforce-hosted-mcp-servers.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
