HubSpot Claude Connector: The Complete Guide

The HubSpot MCP server enables secure read and write access to your HubSpot CRM data - contacts, deals, engagements and more. Once Claude is connected, an AI request becomes a real action against your CRM rather than a description of one. HubSpot documents the connector officially across its knowledge base and developer docs.

• Create and update CRM records - contacts, deals and companies - from the Claude chat window.
• Log activities - notes, tasks and follow-ups - directly on the right records.
• Move deals through pipeline stages and update properties.
• Search and retrieve records with natural-language queries.
• Generate insights from engagement history.

• Post-call admin: "Log this call summary on the Acme deal, create a follow-up task for next Tuesday." Done.
• Pipeline reviews: "Show me all active deals over £50k by stage and last activity" - and Claude summarises.
• Lifecycle tracking: ask Claude to surface contacts that have been sitting in a lifecycle stage for too long.
• Email-driven follow-ups: create follow-up tasks automatically when a key email is opened.
• Campaign analysis: analyse campaign revenue and summarise active deals organised by amount.

Reception across community and partner guides (Speakeasy, Composio, Merge.dev, Windsor.ai, SyncGTM, Scopious Digital) has been notably positive, with the consistent praise being how quickly busywork compresses: post-call notes, pipeline updates and basic reporting become natural-language asks instead of clicks-per-record.

The honest counter-weight: this is a write-capable connector against your live CRM. Bad prompts can produce bad records. Start with retrieval and reporting, then layer in writes once you've validated behaviour on a small set.

HubSpot has had a comprehensive REST API for years - CRM objects, associations, engagements, properties, the lot. If you are building a product integration or a scheduled sync, the API is still the right tool: you control auth, rate limits, retries and schema validation, and you can wrap it in your own business logic.

The MCP server is a different shape. It is built for an agent rather than a program. Instead of you mapping CRUD calls to endpoints, Claude is handed a set of tools - read CRM records, search, create and update records, log activities - and it picks the right one from a natural-language request. The MCP layer also handles OAuth, respects the signed-in user's HubSpot permissions automatically, and exposes objects (contacts, companies, deals, tickets, line items, products, quotes, invoices, orders, carts, subscriptions and more) without you writing a line of endpoint code.

Rule of thumb: deterministic, high-volume, scheduled work belongs on the API; ad-hoc, conversational, "just deal with this for me" work belongs on the MCP server. Most teams end up using both.

HubSpot's remote MCP server went generally available in April 2026 at https://mcp.hubspot.com. It is the cleanest path for most people because the auth is handled for you - you connect any OAuth 2.1 client and sign in with your normal HubSpot account.

Option 1 - Remote MCP server (recommended)

Add the hosted endpoint to Claude Code and complete the OAuth handshake in the browser:

claude mcp add --transport http hubspot https://mcp.hubspot.com

Then run /mcp inside Claude Code, choose HubSpot, and authenticate. To wire the same endpoint into Claude Desktop, go to Settings → Connectors → Add custom connector and paste the URL - Desktop drives the OAuth flow for you.

If you are a HubSpot developer building your own client, create an app under Development → MCP Auth Apps in your account. HubSpot mints OAuth credentials (client ID and secret) automatically, and any OAuth + PKCE client can then connect. The developer MCP requires Developer Platform v2025.2.

Option 2 - HubSpot CLI

hs mcp setup

A guided local setup in under five minutes - useful if you prefer a stdio server tied to the HubSpot CLI you already use.

Option 3 - Open-source npm package

Install the community mcp-hubspot package and configure it in Claude Code if you want to self-host and audit the source. All three paths reach the same underlying CRM capabilities; pick the one that matches your stack.

The HubSpot connector automatically respects user permissions defined in HubSpot - so individual reps only see and act on records they already have access to. That's a meaningful boundary: the agent inherits your HubSpot identity, no elevation. Beyond that, scope what the connector can do, prefer read for sensitive properties, and audit a sample of writes during pilot.

• Claude can't see a record you can see in HubSpot: the connector inherits your HubSpot permissions, but it also depends on the OAuth scopes granted during connection. If a whole object type is missing, disconnect and reconnect so the scope set is refreshed.
• OAuth completes but no tools appear: almost always the browser handshake didn't finish - run /mcp, remove HubSpot, re-add it, and make sure you land back on the "connected" screen before closing the tab.
• "Insufficient permissions" on a write: your HubSpot seat lacks edit rights on that object or property. The agent cannot elevate - fix the role in HubSpot, not in Claude.
• Writes land on the wrong record: with common company names Claude can match the wrong contact or deal. Reference records by a unique attribute (email, deal ID) in your prompt, and keep the pilot read-first until you trust the matching.
• Plan-gated objects return nothing: objects like quotes, subscriptions or custom objects depend on your HubSpot edition. An empty result can mean "not on your plan" rather than "no data".
• Big searches blow the context window: "show me every contact" pulls far more than Claude needs. Constrain by stage, owner, date or amount so the response stays focused.

The MCP server itself is free - there is no separate HubSpot charge to connect Claude. What you can read and write is governed entirely by your existing HubSpot edition and the permissions on your seat, so some objects and properties only appear on higher tiers. The remote server reached general availability in April 2026.

On the Claude side, the one-click connector requires a paid Anthropic plan; the CLI and npm routes work with whatever Claude Code access you already have. Standard HubSpot API rate limits apply to the underlying calls, so heavy automated use should still go through a purpose-built API integration rather than the conversational MCP path.

• Write impact: records affect real pipeline and reporting - review on rollout.
• HubSpot-shaped: the connector is great inside HubSpot, not a replacement for cross-system orchestration.
• Plan-dependent features: some properties and objects depend on your HubSpot edition.

Sales reps and managers, marketing ops, RevOps and customer success - anyone whose work touches HubSpot regularly. The biggest leverage is for teams currently bottlenecked on CRM hygiene rather than selling.

What does it let Claude do?

Create and update CRM records (contacts, deals, companies), log notes/tasks/activities, retrieve and search, and generate insights.

Is it official?

Yes - documented by HubSpot and listed by Claude as an official connector.

Does it bypass permissions?

No - it inherits your HubSpot permissions.

How fast is setup?

Under five minutes via the HubSpot CLI; the connector path is just a click in Claude.

The HubSpot connector turns CRM hygiene from a daily tax into a side effect of asking Claude for what you need. Start read-first, expand to writes once you trust the behaviour, and let your team get back to selling instead of clicking through records.

For revenue teams, it is one of the highest-leverage connectors in the catalogue. Explore the rest in the complete Claude Connectors Directory.