Your AI assistant is only as useful as the data it can reach.
And for most B2B teams, that data lives in HubSpot — technically accessible through APIs, but not in a way that works naturally with the AI tools you already use every day.
That gap is closing fast. HubSpot now ships its own MCP servers, and the net effect is simple: you can ask plain-English questions of your CRM, and get real answers back.
But there's a catch most posts on this topic miss. HubSpot MCP isn't one product. It's two distinct servers that solve different problems, and picking the wrong one — or more commonly, not knowing both exist — will waste a month of your RevOps team's time.
This guide covers both, what each is good and bad at, and how to think about HubSpot MCP if you run revenue operations on HubSpot in 2026.
MCP stands for Model Context Protocol. It's an open standard — originally proposed by Anthropic — that lets AI assistants securely connect to external systems through a consistent interface.
You can think of it as a USB-C port for AI: instead of every tool needing its own bespoke integration with every AI model, MCP defines a shared plug, and everything that speaks it can talk to everything else.
Before MCP, connecting AI to your CRM meant custom middleware. Someone built an auth layer, wrapped HubSpot's REST API, handled errors, mapped responses into something an LLM could reason about, and kept the whole thing patched. Every new AI tool meant a new integration project.
MCP collapses that. A server exposes a standard set of tools — list_contacts, create_task, get_deal — and any MCP-compatible AI client (Claude, ChatGPT, Cursor, Windsurf, custom agents) can call them. The same server works everywhere. The same permissions apply. The same audit trail exists regardless of which AI made the request.
One framing that helps: native AI connectors — the kind you get when you click "Connect HubSpot" inside Claude or ChatGPT — are like a passkey that unlocks the front door.
HubSpot's own MCP servers are more like a keycard system that controls exactly which rooms each person can enter. Both work. One gives you governance. The other gives you speed. For RevOps teams where audit trails and permission scopes matter, that distinction is not cosmetic.
Most coverage of HubSpot MCP treats it as a developer feature. It is, partly — but the bigger story for RevOps leaders is that it removes a layer of friction that has existed since HubSpot shipped its first API.
Historically, three groups got real value out of HubSpot data: people who knew how to build filtered views, people who could write SQL against a data warehouse export, and developers.
Everyone else submitted report requests and waited. A CRO asking "how many deals over €50k did we lose to competitor X last quarter, and which rep owned them?" would go through a days-long chain of Slack messages before getting an answer.
HubSpot MCP collapses that chain. The CRO asks the question in Claude. The MCP server runs the query. The answer comes back in seconds, with the rows, and with a link back to each deal in HubSpot if they want to drill in.
That matters for three reasons. First, it shifts who can interrogate CRM data from "people who know HubSpot's filter UI" to anyone on the revenue team. Second, because MCP is a structured protocol rather than a chatbot bolted onto HubSpot, every request is predictable, every permission is enforced against the actual user's HubSpot scopes, and every interaction is auditable. You don't get AI going rogue in your CRM — you get AI acting as a faster version of whoever asked it.
Third, and this is the part most people don't think about yet: MCP exposes your CRM's configuration as the ceiling for everything AI can do with it. If your deal stages are a mess, AI summaries will reflect that mess. If your lifecycle stages are inconsistent, AI-generated handoff notes will be inconsistent too. The teams that get the most value from HubSpot MCP are the ones whose HubSpot setup was already clean. This is a point we'll come back to at the end.
If you want to make sure your HubSpot setup is actually ready for AI agents to interact with, talk to Superwork about an MCP-readiness review. It's the most effective place to put an hour of thinking before you wire anything up.
Here's where most write-ups go wrong. They treat "HubSpot MCP" as a single product. In reality, HubSpot ships two distinct servers, and the one you pick depends entirely on what you're trying to do.
This is HubSpot's official, cloud-hosted server. It lives at mcp.hubspot.com, runs in beta, and is the one most people mean when they say "HubSpot MCP."
It lets any MCP-compatible AI tool query your HubSpot CRM using natural language. It supports a solid range of CRM objects out of the box: contacts, companies, deals, tickets, products, invoices, quotes, line items, and engagements. Authentication runs through OAuth with PKCE, which means access is scoped to exactly what the user running the query is already permitted to see in HubSpot. If a sales rep without deal edit permissions asks the AI to update a deal, the request fails — the same way it would if they tried to do it in the UI.
What it's good at: natural-language querying of CRM data. Instead of building a filtered view or commissioning a report, you ask your AI assistant "show me all open deals over €50k in the industrial segment where the last activity was more than 14 days ago" and get an answer in seconds. This is the killer use case. It replaces the 20-minute "build a custom view" dance that happens in every RevOps team every day.
What it's not good at yet: write operations. As of April 2026 the Remote server is read-only. You cannot update a deal stage, create a contact, or log an engagement through it. That limitation will lift, but today it's a real constraint.
Other beta-era quirks worth knowing: scopes are auto-assigned at install, which means as HubSpot adds new tools to the server, users may need to reinstall the connection to pick them up. The server is built on HubSpot's CRM Search API, so there's no semantic or vector search — queries work best when they're reasonably structured.
This is a different tool for a different audience. It's a local server that runs inside your IDE — Cursor, Claude Code, VS Code, Windsurf, and similar — and it gives AI coding assistants HubSpot-specific context they otherwise lack.
You install it with one command (hs mcp setup) using the HubSpot CLI. From that point on, your AI assistant knows the platform you're building on.
It can scaffold new HubSpot projects and UI extensions, generate CMS modules and serverless functions, monitor builds and surface errors inline, and search HubSpot's developer documentation without you ever leaving your editor.
This one is clearly for developers building on HubSpot. If you're a RevOps leader, it's not for you directly — but it matters for how fast your developer partners can ship HubSpot apps, UI extensions, and custom workflows.
The productivity delta between a generic AI coding tool and one that actually understands HubSpot's schema, CLI, and platform conventions is large.
Both of HubSpot's official MCPs require using the latest Developer Platform version, so if you're on an older portal, step one is the platform upgrade.
Two servers, two audiences. Here's a decision framework that fits on one screen.
| If you want to… | Use |
|---|---|
| Let your revenue team query CRM data in plain English | Remote HubSpot MCP Server (beta) |
| Build HubSpot apps, UI extensions, or CMS content faster | Developer HubSpot MCP Server |
| Do both — query the CRM and build on the platform | Install both; they don't conflict |
A few decision rules worth internalizing. If your primary use case is "read CRM data with AI," start with the Remote server.
It's the cleanest permissions model and the lowest setup cost — it's HubSpot-native, so your team doesn't need to learn or trust a third-party middleware layer. This is the one RevOps leaders care about first.
If you're a developer agency or in-house developer building on HubSpot's platform, install the Developer MCP on day one. The ROI is immediate and obvious — it's the difference between your AI guessing at your platform and actually knowing it.
The two servers are not alternatives. A well-run HubSpot shop ends up running both, because they answer different questions: one makes your existing CRM data conversational, the other makes building on HubSpot faster. The install cost of each is low enough that "do both" is usually the right answer.
One limitation worth calling out before you plan around it: the Remote server is read-only today. You cannot have AI update a deal stage, create a contact, or log an engagement through it yet.
That constraint will lift — write support is on HubSpot's public roadmap — but it shapes what you can and can't do in 2026. If you need AI-driven writes against HubSpot right now, you're either waiting for the Remote server to catch up or building a purpose-built workflow with a scoped private app. We'd recommend the former for most teams.
Theory is fine. Here's what the first 90 days of HubSpot MCP look like in practice for a RevOps team — the workflows where it pays for itself fastest.
Pipeline triage at the start of every week. A CRO or head of sales opens Claude on Monday morning and asks: "Give me every deal expected to close this month where the last contact activity was more than 10 days ago, grouped by owner."
The MCP server runs the query. The AI returns a prioritized list with deal values, owners, and the last activity date. The whole thing takes 30 seconds. The same question used to mean a custom view, a screenshot in Slack, or a scheduled report that nobody read.
Account briefs before every customer call. Fifteen minutes before a meeting, the AE asks the AI: "Summarize the last 90 days of activity on the Acme account — every note, email, deal update, and support ticket."
Instead of tab-hopping between the company record, the associated deals, the ticket queue, and their email, they get a clean two-paragraph summary with links. Pre-call prep goes from 20 minutes to 2.
Overdue task and follow-up sweeps. A sales manager asks: "List every overdue task across the team, and group them by rep." The MCP server pulls the data. The AI flags the reps with the most slippage. The conversation that follows with each rep is grounded in specifics rather than vibes.
Deal summarization for forecast reviews. Before the weekly forecast meeting, the CRO asks for a one-line summary of every deal over €25k in commit and best-case, including the latest signal from engagement data. What used to be a VP-of-Sales homework assignment becomes a 90-second prompt.
CRM hygiene checks. The RevOps lead asks: "Find every open deal without a close date, every contact without a lifecycle stage, and every company with more than 5 associated contacts but no primary contact flagged." The answers come back as a worklist. Cleanup happens in hours instead of a Q3 project.
Handoff notes from sales to CS. When a deal closes-won, the AE asks the AI to generate the handoff doc: decision criteria from the deal notes, the buying committee from associated contacts, the procurement process summary, and the original reason for the project. CS starts onboarding with context that used to take a 45-minute meeting to transfer.
None of these are science fiction. All of them are running in real RevOps teams today on the Remote HubSpot MCP plus a cleaned-up HubSpot configuration. If you want to figure out which two or three of these would have the biggest impact on your team specifically, that's the kind of thing a conversation with Superwork is built for.
HubSpot MCP is genuinely useful. It is also in beta, and there are things worth knowing before you wire it up to your primary portal.
Start in a sandbox. HubSpot lets you build a standard sandbox from Settings → Account Setup → Sandboxes. Do that first. Connect your AI client to the sandbox portal, test your workflows there, confirm the permissions model behaves how you expect, and only then point at production. This is five minutes of setup that prevents the kind of incident that ends up in a Monday morning all-hands.
Manage scopes deliberately. Scopes on the Remote server are auto-assigned at install — which means the AI gets access to whatever the installing user has access to. If a sales rep with Super Admin rights installs the connection for the whole team, the AI inherits Super Admin rights on their behalf. Have the RevOps lead or a restricted service user install it, not the CRO with full access.
Protect sensitive fields. HubSpot's Sensitive Data feature matters more than ever in an MCP world. If you store PII, financial data, or health data in custom properties, mark them as sensitive and verify your MCP setup respects that flag. AI tools will happily include sensitive fields in responses unless the server is told not to.
Audit, don't trust. Every MCP server should log every request, and HubSpot's official servers do. Set up weekly review of the MCP audit log for the first month — you'll learn a lot about how your team actually uses it, and you'll catch any unexpected query patterns early.
Set expectations with your team. The Remote server is beta. It will change. Tools will be added, scopes will shift, the UI in HubSpot will evolve. Tell your team this is a live system, not a locked-down product — and that their feedback in the first 60 days is how you find the use cases worth investing in.
The direction of travel is clear. The Remote HubSpot MCP server will open up write access. Agentic workflows — where AI doesn't just answer questions but takes multi-step actions on your behalf — will become normal rather than experimental. The line between "querying your CRM" and "operating your CRM" will blur.
When it does, one thing becomes the constraint: your HubSpot configuration quality.
AI can only reason about what exists in your CRM. If your pipeline stages are defined with discipline, AI can give you accurate forecasts. If they're not, AI will generate confident-sounding forecasts that are wrong. If your lifecycle stages are consistent, AI can generate useful lead routing. If they're inconsistent, AI will generate useful-looking routing that sends SQLs to the wrong reps.
The teams that win the next two years of AI-in-CRM aren't the ones who adopt HubSpot MCP first. They're the ones whose HubSpot was already clean when they adopted it. Data quality, property governance, stage definitions, and lifecycle discipline — the unglamorous RevOps work — is now the ceiling for everything your AI tools can do for you.
That's a good thing. It means the investment you've already made in getting HubSpot right pays a new kind of dividend. And it means the teams that haven't made that investment yet have a narrowing window to catch up.
HubSpot MCP is not a single product. It's two — the Remote server for querying your CRM, and the Developer server for building on HubSpot's platform. Most revenue teams will start with the first. Most engineering teams building on HubSpot will start with the second. A lot of organizations should run both.
Whichever server you pick, the strategic answer is the same. AI assistants that can talk to your CRM are a meaningful productivity step for revenue teams — not because they replace anyone, but because they collapse the distance between a question and an answer, between an intent and an action. The time savings are real. The governance model is cleaner than the alternatives. And the teams that move first are building the workflows they'll compound on for years.
If you want help figuring out which flavor of HubSpot MCP fits your setup, what your HubSpot configuration needs in order to get real value from it, or how to run a 30-day pilot without creating a data governance problem, that's exactly the work Superwork does. Book a conversation with our team and we'll walk through what a HubSpot MCP-ready RevOps stack looks like for your specific business.
Your CRM is about to get a lot easier to talk to. The question is whether what it has to say is worth hearing.