Skip to content
Book a build

HubSpot AI: what Claude, ChatGPT and Breeze can (and can't) configure

HubSpot AI compared: what Claude, ChatGPT, Breeze, HubSpot's MCP server and Agent CLI, and Jetstack AI can read, create and change in a portal.

Jetstack AI / Working guides

The AI capability table

Know what each tool can change

Capability table
AssetClaudeChatGPTBreeze
RecordsEditEditSome
PropertiesCreate——
WorkflowsNoNoAgent Hub

Checked against HubSpot's docs · Build the setup in Projects

On this page

Updated November 5, 2026. Checked against HubSpot’s Knowledge Base and developer docs on October 29–30, and against HubSpot’s Agent CLI guide on November 5.

Every AI tool for HubSpot can read your data, and most can change records and marketing content. Very few can configure the portal itself. Claude and ChatGPT can’t build workflows or forms through HubSpot’s connectors, Breeze builds workflows only inside HubSpot’s workflow builder, and HubSpot’s MCP server creates properties and pipelines but has no workflow tool. The exception is HubSpot’s Agent CLI, in beta since June 2026, which lets Claude Code, Claude Cowork and Codex create workflows from JSON.

Which HubSpot AI tool can configure what?

The answer depends on the asset, not the model. Claude, ChatGPT and HubSpot’s MCP server all call HubSpot’s own tools, so they share most of the same gaps. Breeze works inside HubSpot’s editors. Jetstack AI works on setup rather than records.

AssetClaude connectorChatGPT appBreeze Assistant / Agent HubHubSpot MCP serverJetstack AI
CRM recordsRead, create, update; 10 at a timeRead, create, update; 10 at a timeSummarizes records; creates deals, notes and productsRead, create, updateNo
PropertiesRead, create; no updateRead, create; no updateNot describedRead, create; editing not documentedProjects builds them, including calculated, rollup and synced
PipelinesRead, create, updateRead, create, updateNot describedCreate and manageProjects builds deal and ticket pipelines
WorkflowsNo; Claude drafts, you buildNot listedBuilds and edits inside the workflow builderNo workflow toolProjects builds them, switched off
FormsNot listedNot listedNot describedNo forms toolProjects builds them
Lists (segments)Read, create, updateRead, create, updateFinds matching records; creating lists not describedCreate; add or remove records in static segmentsProjects builds them
Marketing emailsRead, create, updateRead, create, updateRefines content in the editorRead; create and edit draftsProjects builds them
Custom objectsRecords onlyRecords listed; HubSpot’s FAQ says not available yetNot describedRecords; object definitions not documentedProjects builds them
Reports and dashboardsSaved reports, read onlyAnalytics, read onlyGenerates charts from your dataAnalytics, read onlyProjects builds them, through the Chrome extension

“Not listed” means HubSpot’s article for that tool doesn’t include the asset. “Not described” means HubSpot’s Breeze articles don’t mention it. Neither is a promise that it will never work, so test anything you plan to rely on.

Deletes. HubSpot lists no delete actions for its Claude or ChatGPT connectors and documents no delete tool for its MCP server. One exception showed up when we checked on October 29, 2026: the live Claude connector’s segment tool offered to delete and restore lists.

HubSpot’s Agent CLI (beta). A separate command-line tool for AI agents such as Claude Code, Claude Cowork and Codex, released in public beta in June 2026. It reaches more configuration than the connectors: workflows (create and update from a JSON definition), properties, pipelines and stages, custom object schemas, association labels and reports, plus records. Your agent writes the JSON and works out the order, and every write can be previewed with --dry-run. HubSpot’s guide names forms and lists but documents no commands for them. See the HubSpot Agent CLI guide.

Announced, not yet documented. HubSpot’s Fall 2026 developer changelog says its MCP server can “create and manage” custom properties and schedule and publish marketing emails. The tool reference still describes creating properties and editing drafts.

For Claude in detail, including setup for each route and the full permission model, read the Claude HubSpot integration guide.

What each tool needs before it works

ToolHubSpot planOther requirementsWho approves
Claude connectorAll plansPaid Claude plan: Pro, Max, Team or EnterpriseSuper Admin, for users without App Marketplace access
ChatGPT appAll plansAn OpenAI accountSuper Admin, for users without App Marketplace access
Breeze AssistantAll plans; workflows and agents need Professional or EnterpriseAn admin turns on Breeze Assistant access in AI settings; Agent Hub agents use HubSpot CreditsYour user permissions decide what it can do
HubSpot MCP serverAny; some tools need Professional or EnterpriseAn MCP client that supports OAuth with PKCEThe connecting user
Jetstack AIA portal connected to Jetstack AIProjects is in limited release; the MCP server needs a Jetstack access tokenToken scopes, and separate approvals in each Project

HubSpot’s Claude and ChatGPT connectors act as the connected user, and their creates and updates appear in HubSpot’s Audit Log under both the user and the connector.

Can Breeze build HubSpot workflows?

Yes, inside HubSpot. In Agents → Agent Hub, open the Workflows tab, select or create a workflow, and click the AI icon at the bottom of the workflow builder. Breeze Assistant adds triggers and actions to the canvas, configures workflow settings and can generate the code for a custom code action. You need a Professional or Enterprise subscription and Super Admin or Workflows permission.

Breeze works on the one workflow in front of you. HubSpot’s articles don’t describe it creating the properties, lists, forms or pipelines that workflow depends on, so those have to exist first. Its changes appear on the canvas straight away, so review them there before you turn the workflow on.

Can Claude or ChatGPT build HubSpot workflows?

Not through HubSpot’s connectors. Neither connector’s article lists workflows, HubSpot’s MCP server has no workflow tool either, and HubSpot’s Claude article says Claude drafts workflows for you to build. In Claude Code, Claude Cowork or Codex, HubSpot’s Agent CLI (beta) creates, updates and deletes workflows from a JSON definition, and a developer can also call HubSpot’s beta Automation API directly. Either route makes you responsible for the portal’s real IDs, the order the dependencies are created in, and the API’s update rule: anything left out of an update is removed.

What we recommend for building

Use the AI assistants for what they do well: answering questions about your data, drafting content and updating records with a confirmation each time. For new setup, such as properties, pipelines, lists, forms, workflows and the reports that track them, we recommend Jetstack AI Projects.

You describe the work in a brief and chat. Jetty, the Projects assistant, turns it into a plan you review, and approving the plan starts generation. You review the generated assets and approve deployment separately. A Project is bound to one portal and builds in dependency order: properties before the workflows that use them. Workflows are created switched off, and some assets go live through the Chrome extension. After the run, you get a receipt of what was created, checked by reading it back from HubSpot.

Projects can change only assets it built: properties, pipelines, active lists, reports and switched-off workflows. For something built elsewhere, it builds a corrected replacement, created switched off, and you retire the old one in HubSpot. It doesn’t build CMS pages, themes or custom-code actions, and it doesn’t import data, manage users or set up integrations. Projects is currently in limited release; see what it can build and Projects.

How it compares in each case: Jetstack vs Claude and the HubSpot MCP, Jetstack vs ChatGPT and Jetstack vs Breeze.

How to bound an AI task in HubSpot

Whichever tool you use, these steps keep a write from turning into a cleanup.

  1. Separate advice from execution. Drafting a campaign or explaining a workflow needs read access at most. Creating connected setup needs write access, a dependency plan and a way to check the result.
  2. Check the actual connection. Confirm which portal is connected, which user authorized it and which data permissions were approved. Finding a contact doesn’t mean the same connection can edit a property or a workflow.
  3. Write the task with exclusions. Name what it may read, what it may change, and what it must not touch, such as “don’t activate workflows, don’t delete anything, don’t contact customers”.
  4. Keep approvals on. HubSpot recommends setting the Claude connector’s write tools to need approval. Read each proposed change before you say yes, and don’t waive confirmations on a production portal.
  5. Read the results per asset. Ask which assets were created, changed, skipped or failed. HubSpot’s property and pipeline tools return a result per item, so one can fail while the rest succeed.
  6. Inspect before you retry. A timeout isn’t proof nothing was written. Check the portal before running the same create again.
Before you let an AI tool change HubSpot

A reading checklist. Selections stay on this page only and are not saved.

Sources

The implementation workflow

Know what each tool can change.

A dated view of what Claude, ChatGPT and Breeze can configure, and where to build instead.

Explore Projects