If you’ve seen “HubSpot MCP” in an AI tool’s docs or a LinkedIn post and wondered what it actually means, here’s the short version as of April 2026: it primarily refers to HubSpot’s own Remote MCP Server, which went generally available on 2026-04-13 and lives at https://mcp.hubspot.com.

This post explains what that covers, how it compares to the ChatGPT connector and the Developer MCP Server, and where third-party MCPs still fit. For the full deep reference, see the HubSpot MCP pillar guide.

MCP is a protocol, not a product

MCP stands for Model Context Protocol - an open standard for connecting AI assistants to external tools and data. Anthropic released it in late 2024, and every major AI client (Claude Desktop, Cursor, VS Code, OpenAI’s Codex CLI, Gemini CLI, ChatGPT via the HubSpot connector) now speaks it. Think of it as the USB-C of AI tooling: one plug, many compatible devices.

“HubSpot MCP” is shorthand for “an MCP server that talks to HubSpot”. HubSpot now ships the first-party one.

HubSpot’s Remote MCP Server is the main thing to know

The Remote MCP Server is first-party, HubSpot-hosted, and free to all HubSpot accounts across all hubs and tiers. At GA on 2026-04-13, it supports:

• Read and write on 12 CRM object types: contacts, companies, deals, tickets, carts, products, orders, line items, invoices, quotes, subscriptions, and segments (lists).
• Read and write on all 5 engagement types: calls, emails, meetings, notes, tasks.
• Read-only access to organizational context (users, teams, reporting structures, owners, roles, seats) and marketing content (campaigns, landing pages, website pages, blog posts).

All actions respect the connected user’s existing HubSpot permissions. A rep can only see or modify records they already have access to inside HubSpot. The server exposes a fixed set of 12 tools - the important ones are get_user_details (call it at session start to check what’s available), search_crm_objects, get_crm_objects, and manage_crm_objects (the write path).

What it doesn’t do today. Custom object schemas are not exposed yet. There’s also no vector/semantic search - the server is built on HubSpot’s CRM Search API, so queries are filter-based. And if your account has Sensitive Data enabled, engagement objects get blocked from MCP access (CRM objects are unaffected).

See the full pillar guide for the exact limits (get_crm_objects 100 IDs per call, search_crm_objects 200 results per page, 5 filter groups × 6 filters, 5 keywords per property search), the 12 tools in one scannable list, and the step-by-step OAuth 2.1 + PKCE setup flow.

The ChatGPT connector is the Remote MCP with a different install path

The HubSpot Connector for ChatGPT (announced 2026-02-26) is a managed install that lives inside ChatGPT itself. It uses HubSpot’s MCP machinery underneath, but the install flow and some of the constraints are different:

• A ChatGPT workspace admin approves the connector for the workspace; individual users upgrade and re-authenticate.
• Write actions have a 10-record bulk cap per request (the general Remote MCP doesn’t have this cap in the same form).
• Attribution is recorded to both the user and the ChatGPT connector in HubSpot’s audit log.
• Same Sensitive Data caveat - engagements blocked if the setting is on.

Pick it if your team lives in ChatGPT Business/Enterprise and you want HubSpot available in every chat without running an MCP client yourself. Pick the direct Remote MCP if you’re using Claude Desktop, Cursor, or VS Code, or if you want to avoid the 10-record write cap.

Don’t confuse it with the Developer MCP Server

HubSpot ships two MCP servers, and the names are confusingly similar. The Developer MCP Server is a local, CLI-installed product (hs mcp setup) for developers building HubSpot apps and CMS content inside tools like Claude Code, Cursor, VS Code, and Gemini CLI. It went GA on 2026-02-19.

If a RevOps team is asking about “connecting Claude to our HubSpot CRM”, the answer is the Remote MCP Server. If a developer is asking about “AI-assisted HubSpot app development in Cursor”, the answer is the Developer MCP Server. Different products, different endpoints.

Where third-party MCPs still fit

The Remote MCP Server covers most of what a RevOps team needs. But there are real gaps where third-party MCPs are still the right fit for specific jobs.

• Custom objects. The Remote MCP doesn’t expose custom object schemas today. Daeda AI’s Automation Suite supports existing custom objects - sync, query, add properties, create records, update records. It connects your AI assistant to a Daeda-managed database service, authenticates via HubSpot OAuth with deep optional scopes, and builds writes as human-approved Write Plans you review and confirm in one step. (Creating brand-new custom object schemas from a third-party app is blocked by a HubSpot platform scope that isn’t released yet.)
• Local, sub-second querying on contacts, companies, and deals. If you’re a RevOps power user and you want your AI to answer questions about those three objects without a network round-trip every time, the free daeda-mcp npm package syncs them to an encrypted SQLite database at ~/.daeda-mcp/data/ on your own machine. It’s strictly scoped - contacts, companies, and deals only, read-only, HubSpot private-app token auth - but it’s purpose-built for that workflow.
• Multi-portal agencies. One OAuth connection to mcp.hubspot.com equals one portal. Agencies running many client portals hit reconnect-fatigue fast. Daeda AI’s Automation Suite has a one-workspace-many-portals model - switch between connected portals from inside your AI assistant without re-authing.
• AI actions inside HubSpot workflows. MCP is for external AI clients talking to HubSpot, not for dropping AI into a HubSpot workflow step. Daeda AI’s Ask AI workflow action runs inside HubSpot workflows with a BYOK OpenRouter key.
• Human-approved Write Plans. The Remote MCP writes immediately when you ask. Daeda AI’s Automation Suite builds each change as a Write Plan that a human reviews and approves inside Daeda AI in HubSpot before it runs.

None of those are “the Remote MCP is bad.” They’re “here’s where the decision matrix forks.”

Where to start

• If you just want HubSpot + your AI client working this week, set up the Remote MCP Server. HubSpot’s setup guide walks you through the MCP Auth App creation and the OAuth 2.1 + PKCE flow.
• If you want sub-second local queries on contacts, companies, and deals, install the free daeda-mcp npm package and point Claude Desktop or Cursor at it.
• If you need custom objects, multi-portal, in-workflow AI actions, or human-approved Write Plans, look at Daeda AI on the HubSpot Marketplace. 7-day free trial.

The HubSpot MCP pillar guide has the complete decision tree, the exact tool list, the real limits, and the Sensitive Data caveat in full. Read that if you want to go deeper on any of this.