Most of the coverage of HubSpot’s recent product releases focused on the parts visible in the UI — the new Breeze Assistant interface, the Context Home dashboard, the Marketing Studio campaign canvas. Those features are real and worth understanding, but they’re surface-level compared to something HubSpot shipped alongside them that got almost no attention: the HubSpot MCP Server.
If you don’t know what MCP is, that’s most marketers’ starting position. This post will fix that, because MCP is likely to matter more for how AI tools interact with your CRM over the next few years than any of the in-platform AI features HubSpot has shipped so far.
What MCP Is
MCP stands for Model Context Protocol. It’s an open standard developed by Anthropic that defines how AI models can connect to external tools and data sources in a structured, secure way.
Before MCP, getting an AI model to work with your company’s data required custom integrations — you’d export data to a file, paste it into a prompt, or build a bespoke API connection that the AI could query. Each integration was one-off, required maintenance, and was scoped to a specific use case.
MCP standardizes this. An AI tool that supports MCP can connect to any MCP server through a common protocol. The server exposes a defined set of capabilities — data it can query, actions it can take — and the AI tool can discover and use those capabilities without custom code being written for each specific combination.
Think of it like USB for AI tools. Before USB, every device needed its own cable and driver. USB created a standard connection that any device could use with any computer. MCP does the same thing for AI model connections to external data and tools.
What the HubSpot MCP Server Exposes
The HubSpot MCP Server is a server you can run that gives MCP-compatible AI tools access to your HubSpot portal. What it exposes depends on the implementation, but the current capabilities include:
CRM data queries. You can ask an AI tool to retrieve contact records, company records, deal information, and pipeline data from your HubSpot portal. This includes custom properties you’ve defined, lifecycle stages, and associated activity records.
Contact and company search. The MCP server allows AI tools to search your CRM by name, email, domain, or custom property values. A request like “find all contacts at companies with more than 200 employees that are in the consideration stage” is executable through the MCP connection.
Deal and pipeline access. Deal records, associated contacts, current stage, close dates, and deal amounts are all queryable through the MCP server. This is the foundation for building deal intelligence workflows that work outside the HubSpot UI.
Activity and engagement history. Email opens, form submissions, meeting logs, and call records associated with a contact or deal can be retrieved through the MCP connection.
Property updates. Some implementations of the HubSpot MCP Server allow write access — the ability for an AI tool to update CRM properties based on instructions. This is where it gets powerful and where the permissions model matters.
Why This Is Different From the HubSpot API
HubSpot has had a REST API for years, and sophisticated teams have been using it to build custom integrations. What makes the MCP server different is the audience and the interaction model.
The HubSpot API requires a developer to write code. You define what you want, you write the query, you handle the response format, you manage authentication. It’s powerful but requires engineering time for every new use case.
The HubSpot MCP Server is designed to be queried by AI models in natural language. Instead of writing a query, you describe what you want in plain language to an AI tool like Claude, and the AI model figures out how to retrieve it from the HubSpot MCP server. No code written, no query structured — just a description of the information you need.
This is the fundamental shift. The MCP server makes HubSpot data accessible not just to developers who can write queries, but to anyone using an MCP-compatible AI tool who can describe what they need.
What You Can Actually Build
Custom deal review agents. Connect Claude or another AI model to your HubSpot MCP server and ask it to run a weekly deal review: “Look at all open deals in the closing stage that haven’t had a logged activity in 14 days. Draft a brief status summary for each one and flag which ones need rep action.” The AI queries your HubSpot data, synthesizes it, and returns actionable output without you navigating the HubSpot UI.
Contact research automation. Before a sales call, ask an AI tool to pull everything HubSpot knows about a contact and their company, summarize recent engagement history, and draft a brief call prep document. This is a task that currently takes a rep 10-15 minutes manually and produces inconsistent results. With an MCP connection, it’s a 30-second prompt.
Anomaly detection workflows. Ask an AI model to monitor your pipeline for patterns that indicate forecast risk — deals where engagement frequency has dropped, deals that have been in the same stage longer than your median, contacts that have opened emails but aren’t responding to follow-up. The AI can pull this data from your HubSpot MCP server, analyze it against historical patterns, and surface the outliers.
Cross-tool synthesis. If you also have MCP servers set up for other tools in your stack, you can ask AI models to work across multiple data sources simultaneously. Pull deal context from HubSpot, conversation intelligence from your call recording tool, and intent signals from your data enrichment provider, then synthesize them into a single briefing. This kind of cross-tool synthesis is what most RevOps teams do manually in weekly reviews — an MCP-connected AI model can do it in seconds.
Custom reporting on non-standard questions. HubSpot’s built-in reporting is strong but constrained to the reports HubSpot has designed. With an MCP connection, you can ask non-standard questions: “How many deals where the first contact was inbound converted to closed-won in under 30 days, broken down by industry?” This is a specific query that HubSpot’s report builder may not support directly, but an AI model with MCP access can construct and answer it.
Who This Is Actually For
The HubSpot MCP Server is not a feature for everyday marketing teams. It’s infrastructure for technical users — RevOps professionals, marketing engineers, and developers building internal tools. Getting it set up requires running a server, managing authentication, and configuring what the server exposes to connected AI tools.
That said, the audience for this capability is expanding. MCP-compatible AI tools are becoming more accessible, and the configuration overhead is decreasing as more turnkey implementations become available. Teams that have a RevOps function or access to a technically capable marketing operations resource can get value from this today.
For everyone else, the more immediate relevance is understanding that MCP is the direction HubSpot’s AI integration model is heading. The in-platform AI features — Breeze Assistant, Context Home, Agent Hub — are the consumer-facing layer. MCP is the infrastructure that will eventually power more sophisticated use cases built on top of your HubSpot data.
Setting Up the HubSpot MCP Server
The HubSpot MCP Server is available as an open-source project. At a high level, the setup involves:
Step 1: Get your HubSpot API credentials. You’ll need a private app token from your HubSpot developer account with permissions scoped to the data you want to expose. Only grant the permissions you need — read-only access to CRM objects is the right starting point for most implementations.
Step 2: Install and configure the MCP server. The server runs as a Node.js application. You configure which HubSpot objects it exposes and what operations are permitted — query only, or query and write.
Step 3: Connect your AI tool. MCP-compatible AI tools like Claude support MCP server connections through their configuration. You add the server’s connection details, and the AI tool discovers the available capabilities automatically.
Step 4: Define what the server exposes. Most teams start with read-only access to contacts, companies, and deals. If you want write access, add that only after testing read operations and confirming the AI tool is querying and interpreting your data correctly.
Step 5: Test with specific queries. Before using MCP-connected AI tools for anything that affects your data, spend time on read-only queries to confirm the responses are accurate. Check edge cases — contacts with unusual data, deals in non-standard stages — to make sure the AI is interpreting your CRM structure correctly.
The Permissions and Security Question
The most important thing to get right with an MCP server connection is the permissions model. An AI tool with write access to your HubSpot CRM can update records at a speed and scale that a human can’t match — which means a misconfigured instruction or a prompt that’s interpreted differently than intended can make changes across your CRM before anyone notices.
The standard recommendation is to start with read-only access and add write permissions only for specific, well-defined operations where the blast radius of an error is limited. Updating a single custom property is lower risk than bulk-updating lifecycle stages.
Review HubSpot’s developer documentation for the current scope options and follow the principle of least privilege — only expose what your actual use case requires.
What You Can’t Do With It Yet
The HubSpot MCP Server doesn’t give AI tools access to everything in your HubSpot portal. Marketing Hub campaign data, reporting, and workflow logic are either partially or not yet exposed through the current MCP implementation. The server is primarily CRM-layer access — contacts, companies, deals, and associated activities.
As HubSpot expands its MCP implementation, expect broader coverage of Marketing Hub and Service Hub data. The CRM-first approach makes sense as a starting point: that’s where the highest-value queries live for most teams.
For teams currently using Agent Hub’s built-in agents, the MCP server is additive, not competitive. The in-platform agents work within HubSpot’s own workflow engine. The MCP server extends access to any external AI tool that supports the protocol. They serve different use cases and can run alongside each other.
HubAISpot covers independent, practical guides on HubSpot and AI for marketers and sales teams. No vendor relationships. No paid placements.
Leave a Reply