HubSpot Builds AI-Friendly Backend, Connecting Tools Like Claude and Cursor to CRM Data

AI-generated image of interconnected wires plugging into computer ports. Image credit: Adobe Firefly

HubSpot has opened up a Model Context Protocol (MCP) server in public beta, enabling developers to tie their AI apps with CRM data. The company states that this offering is an NPM package and is “intended” for developers, technical teams, and companies that want to build custom apps or integrations leveraging large language models (LLMs).

All Your HubSpot Data Are Now In AI

MCP is a standard that allows AI models to connect with applications through a consistent interface. As HubSpot describes it, MCP “acts as an abstraction layer over traditional APIs, letting AI agents access application functionality without needing to understand specific API protocols.” So, like how Google, Facebook, and Salesforce have API feeds from which developers can connect their software to those platforms, MCPs perform the equivalent function in this AI era.

With this access, AI apps like Anthropic’s Claude and Cursor can access HubSpot directly to inform prompt responses. Through integrated third-party programs, users can retrieve, create, or modify HubSpot objects, list them and their properties, create tasks and notes within their accounts, and provide feedback on the MCP server using natural language.

Here’s an example the company provided when asking AI to streamline development:

“Get the company record for Google and place it in a JSON file that I can reference in my code as a placeholder as work on the UI for my UI Extension, website module, and template.”

Alternatively, these prompts could be used in apps to get insights or create CRM records:

“Summarize all deals in the “Decision maker bought in” stage in my HubSpot pipeline with deal value > $1000.”

“Update the address for John Smith in my HubSpot account”

HubSpot discloses that accessing the MCP server will require some technical proficiency. At the very least, you’ll have to edit a JSON file and create a HubSpot Private App. In addition, Node.js must be installed.

Helping SMBs Play Big

By opening an MCP server, HubSpot equips its small—and medium-sized customers with tools that allow them to operate with the sophistication of enterprise businesses. The company told me in April that its mission is to empower SMBs to leverage AI like their much larger counterparts, but without the need for hefty budgets and resources.

“SMBs are constantly struggling with, ‘I don’t have enough people. I don’t have enough money. I don’t have enough expertise,” Nicholas Holland, HubSpot’s head of AI and its senior vice president of product, explained in an interview. “We’re bringing AI to the table to help solve these issues in a way that…a large company would throw money or people at it. We want to bring this to something where it’s not the size of the money that determines success.”

Working with AI apps isn’t exclusive to the enterprise. SMBs also want to play in this arena, preferring to develop AI that incorporates their HubSpot data so the responses are relevant to their needs. This is another step to reduce the so-called “cobble tax,” a penalty SMBs face for having to authenticate across different tools. Now, SMB customers won’t need to sort out a workaround to get their CRM data into AI—it’s now directly connected through the MCP server.

All this being said, HubSpot isn’t the only CRM provider with an MCP server. It joins Zoho, Pipedrive, and others in granting AI apps access to their data via this standard.

How will this offering affect HubSpot’s own AI tools, Copilot and Breeze? The company says that while third-party apps may outperform in certain use cases, there will likely be scenarios and tasks where its native tools have the best experience.

Featured Image: AI-generated image of interconnected wires plugging into computer ports. Image credit: Adobe Firefly

Leave a Reply

Discover more from Ken Yeung

Subscribe now to keep reading and get access to the full archive.

Continue reading