Your CRM, inside every AI tool.

Until recently, every AI tool lived on an island. Staff copied customer data into a chat window, got an answer and pasted it back. That was slow and risky. MCP replaces the copy and paste with a governed connection, so the AI tool your people already use can work with live company data.

What is MCP, in plain terms?

Think of it as a standard plug. Any system that offers an MCP server can connect to any AI tool that speaks MCP. In December 2025, Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI and backed by Google, Microsoft, AWS and others. At that point, the MCP project reported more than 97 million monthly SDK downloads. It is now neutral infrastructure, not one vendor's feature.

What does HubSpot offer?

HubSpot provides connectors for Claude, ChatGPT, Gemini and Copilot, and states that its remote MCP server is generally available, letting any MCP-compatible AI tool read and write to the HubSpot CRM through conversation, with “permission-aware access”. A sales leader can ask which deals are at risk this quarter, or update a record from meeting notes, without opening the CRM.

Why does this matter to a CEO?

Because it changes where value sits. If every AI tool can reach the CRM, the choice of AI tool matters less, and the quality of the customer record behind it matters more. A clean, connected HubSpot becomes the context every assistant draws on. A messy one feeds every assistant the same mistakes.

What are the risks?

Write access is the big one. An assistant that can update records can also update the wrong ones. Decide which teams may connect which tools, whether access is read only or read and write, and who reviews the list of connections. Under Malaysia's amended PDPA and similar laws in the region, you also need to know where each AI provider processes data.

Do we still need HubSpot's own agents?

Yes, for different jobs. Connectors serve people asking questions in their favourite assistant. Agents in Agent Hub run work continuously inside HubSpot, such as resolving tickets or prospecting. Most enterprises will use both, on the same customer record.

Where should you start?

With one team and one question they ask every week. For a sales leadership team, that might be which deals moved, stalled or slipped since Monday. Connect the approved assistant to HubSpot with read only access, and let that team use it for a month. Watch what they ask, what they trust and where the answers fall short. The gaps will almost always point to the data: missing fields, duplicate companies, notes kept outside the CRM. Fix those, then widen access. This order matters. Connecting every tool to a messy CRM at once spreads poor answers across the company faster than any project could clean them up.

What to do on Monday.

1. Ask IT which AI tools staff already connect to company data, sanctioned or not.
2. Choose the approved assistants and connect them through official connectors, not copy and paste.
3. Start with read only access, then grant write access by team and use case.
4. Add every AI connection to your AI register, with an owner.