# How to Use Claude for HubSpot Workflow Fix Lists

**Quick Answer:** Claude does not fix your HubSpot workflows. It diagnoses them. Connect [Howly's MCP server to Claude Desktop](https://howly.io/#claude), ask a question in plain language instead of clicking through workflows one at a time, and Claude reads your portal's health score, workflow issues, and dependency data through Howly, then returns a prioritized fix list: a ranked breakdown of what is broken, how severe it is, and what to address first. The fix itself still happens by hand, inside HubSpot. Nothing in the chain from Claude to Howly to HubSpot writes to your portal.

%[https://youtu.be/i8iVg1xiHkI?si=afXi15juAn4AGZWs&t=1410] 

* * *

## The Problem With Diagnosing Workflows Manually

A portal with 200 workflows does not give up its problems easily. Something is broken. Contacts are stuck. A property is being set twice by two different workflows, and nobody can say which one is supposed to win.

The instinct is to open HubSpot's workflow list and start clicking. That works for five workflows. It falls apart at fifty, and it becomes a multi-day project at two hundred, because HubSpot's native view shows you one workflow at a time. It does not show you which workflows depend on each other, which ones are stale, or which ones are quietly duplicating logic.

RevOps teams end up with a spreadsheet of findings built by hand, one workflow tab at a time. Marketing ops inherit a portal from an agency and have no starting point at all. Agencies managing a dozen client portals cannot run this process manually for every account without burning hours that were never billed for.

Claude changes the shape of this problem. Instead of clicking through workflows, you ask a question. Instead of building the fix list yourself, you export one.

* * *

## What Claude Actually Does With Your HubSpot Data

Claude does not connect to HubSpot directly. It connects to Howly, and Howly is already connected to your HubSpot portal through a read-only OAuth integration. That distinction matters: Claude is reasoning over data Howly has already mapped, not reaching into your CRM on its own.

Howly's MCP server exposes four tools to Claude, and each one takes specific parameters, which is what makes the follow-up questions work.

*   **Portal health score.** A single 0 to 100 number, plus issue counts broken out two ways: by category (all eight below) and by severity (critical, warning, info). It also returns active workflow count against total, so you can see at a glance how much of the portal is even running.
    
*   **Workflow issues, filtered.** This is the tool most people underuse. You are not limited to "show me everything." You can filter by a specific category, a specific severity, or both, and cap how many results come back. Ask for only critical issues and Claude returns just those, not a wall of low-priority noise mixed in.
    
*   **Workflow list, filtered by status.** Ask for active workflows only, inactive only, or everything. Results come back sorted active-first.
    
*   **Individual workflow detail.** Ask Claude what a specific workflow named "Lead Nurture Sequence" actually does, by name or by id, and it returns the enrollment trigger, every action in order, and any branch logic inside it. If the workflow splits into an if/then, Claude sees inside both branches, not just the top-level steps.
    

This is HubSpot automation analysis done conversationally. You are not reading a static report. You are asking follow-up questions the way you would ask a colleague who already knows the portal, and getting answers grounded in what Howly actually found, not in a general description of what a healthy portal looks like.

### The Eight Issue Categories You Can Filter By

Howly's Health Checker tracks eight specific issue types, and Claude can filter for any one of them directly:

*   **Broken enrollments** (critical): a workflow trying to enroll contacts into a workflow that no longer exists or is turned off.
    
*   **Circular dependencies** (critical): two or more workflows enrolled into each other, forming a loop.
    
*   **Duplicate actions** (warning): the same property set to the same value by more than one workflow.
    
*   **Empty workflows** (warning): a workflow with no actions configured at all.
    
*   **Stale workflows** (warning): active, but not modified in six or more months.
    
*   **Possible duplicate workflows** (info): two workflows with similar names, often a sign of copy-paste duplication nobody cleaned up.
    
*   **Orphan workflows** (info): standalone, with no connection to any other workflow in the portal.
    
*   **Inactive workflows** (info): workflows currently turned off.
    

Knowing this list matters because it tells you exactly what to ask for. "Show me every circular dependency" and "show me only critical issues" are two different, precise questions, and Claude can answer both.

* * *

## Setting Up the Claude AI Integration

Connecting Claude to your HubSpot workflow data takes three steps, and none of them require touching your portal's automation.

**Step 1: Connect Howly to your HubSpot portal.** [Howly is listed on the HubSpot Marketplace](https://ecosystem.hubspot.com/marketplace/listing/howly) as a featured, official Technology Partner app. Authorize the OAuth connection, and Howly loads a full workflow map of your portal in 10 to 25 seconds, depending on portal size. No credentials are shared. No workflow is edited during this step or any step after it.

**Step 2: Connect Howly to Claude Desktop.** From your Howly account settings, generate a personal access token and add it to your Claude Desktop configuration using the snippet Howly provides. This is a one-time setup per workspace.

**Step 3: Ask Claude questions.** Once connected, prompt Claude the way you would prompt any assistant with context. Ask for a health score. Ask what is broken. Ask for a prioritized list. Claude pulls the answer from Howly's live data, not from a cached export.

Agencies managing multiple client portals can repeat this setup per client account, since Howly supports unlimited client portal connections under one agency login.

* * *

## Example Prompts for Automation Troubleshooting

The value of this integration comes from the specificity of what you ask. Vague prompts get vague answers. Specific prompts get a fix list.

**Portal health and priority audits**

*   "What's my HubSpot portal health score right now, and what are my biggest problems?"
    
*   "Give me a prioritized to-do list of everything I should fix in my HubSpot workflows this week."
    

**Redundant and broken workflows**

*   "Which of my workflows are broken and trying to enroll contacts into deleted workflows?"
    
*   "Are any workflows setting the same property to the same value? Find my redundant logic."
    

**Deep-dives and client reporting**

*   "I have a workflow called 'Lead Nurture Sequence.' What does it actually do, including any branches?"
    
*   "Write an email to my client summarizing the current health of their entire HubSpot portal."
    

**Filtered and targeted queries**

*   "Show me only critical issues, nothing lower priority."
    
*   "List every circular dependency in my portal."
    
*   "Which workflows look like possible duplicates based on their names?"
    
*   "Show me every inactive workflow so I know what's safe to archive."
    

Each of these returns something you can act on immediately, not a summary you have to translate into next steps yourself. The filtered prompts matter most once a portal gets large, because "show me everything" stops being useful past a few dozen issues.

* * *

## Turning Claude's Answer Into a Prioritized Fix List

A chat answer is useful in the moment. A fix list is useful for the whole team, which is why the workflow does not end at the Claude conversation.

Ask Claude for the priority ranking directly: "Rank these issues by severity and tell me which three to fix first." Claude will typically separate findings into three tiers, matching how Howly's own AI Audit classifies them: issues that are actively breaking automation, gaps in coverage, and opportunities for cleanup that are not urgent but add up over time.

For the version your team actually works from, run the same audit inside Howly directly. The AI Audit, powered by Claude, completes in about 15 seconds and produces a ranked findings list inside Howly's interface, with one-click links back to each affected workflow in HubSpot. Agencies can export that audit as a branded, white-label PDF report for a client handoff or a quarterly review. This is the same underlying automation troubleshooting logic Claude gives you in chat, packaged as a document your client or your VP of Marketing Operations can actually open.

* * *

## Why the Category Breakdown Matters More Than a Raw Count

A portal health score on its own tells you how bad things are. It does not tell you where to start. This is why the category and severity breakdown matters more than the single number.

A portal with twelve broken enrollments and three stale workflows has a completely different fix list than one with twelve stale workflows and three broken enrollments, even if both land on the same overall score. Claude will name the category directly when it reports a finding, because Howly's underlying data is already organized this way. That specificity is what separates a real fix list from a generic checklist you could have written without looking at the portal at all.

* * *

> **Key Takeaways**
> 
> *   Claude diagnoses and ranks HubSpot workflow issues through Howly's MCP server. It does not edit, fix, or write to the portal. The entire chain stays read-only.
>     
> *   Natural-language prompts replace manual, workflow-by-workflow inspection for portal health, broken enrollments, and redundant logic.
>     
> *   The AI Audit inside Howly, powered by Claude, turns a chat answer into a ranked, exportable fix list in about 15 seconds, with one-click links back to HubSpot.
>     

* * *

## Frequently Asked Questions

**How does Claude access my HubSpot workflow data?** Claude does not connect to HubSpot directly. It connects to Howly's MCP server, and Howly is separately connected to your HubSpot portal through a read-only OAuth integration. Claude reasons over the data Howly has already mapped and cannot write, edit, or trigger anything inside your portal.

**Does Claude fix HubSpot workflows automatically?** No. Claude produces a ranked list of findings: what is broken, how severe each issue is, and a direct link to the affected workflow in HubSpot. Applying the fix, editing a workflow, deactivating it, or rebuilding it, still happens by hand, inside HubSpot. The entire chain from Claude to Howly to HubSpot is read-only, and nothing in this integration writes, edits, or activates anything in your portal.

**Is this different from asking Claude general HubSpot questions?** Yes. A general question to Claude draws on public knowledge about how HubSpot works. A question routed through the Howly MCP integration draws on your actual portal's live health score, workflow list, and issue detection, so the answer is specific to your automation rather than a generic best practice.

**What kind of issues will Claude flag during automation troubleshooting?** The most common findings are broken enrollment chains, duplicate actions setting the same property to the same value, orphaned workflows with no upstream or downstream connections, and stale workflows that have gone six or more months without a modification. Claude reports these using the same categories Howly's Health Checker uses internally.

**Can I get a prioritized fix list instead of a raw list of problems?** Yes. Ask Claude directly to rank findings by severity, or run the AI Audit inside Howly, which classifies every finding as an issue, a gap, or an opportunity and ranks them by priority automatically. The audit completes in about 15 seconds.

**Do I need a HubSpot developer account to set this up?** No. Connecting Howly to HubSpot uses a standard OAuth authorization from your existing HubSpot account. Connecting Howly to Claude Desktop uses a personal access token generated inside your Howly account settings, not a developer API key.

**Can agencies use this across multiple client portals?** Yes. Howly supports unlimited client portal connections under one agency account, and each connected portal can be queried through Claude separately once the MCP integration is set up.

**Will this affect my HubSpot API rate limits?** Howly's connection to HubSpot is designed to stay within standard API limits for portals of any size, and the Claude integration queries data Howly has already retrieved rather than making new HubSpot API calls per question.

* * *

## Summary

HubSpot automation analysis does not have to mean clicking through workflows one at a time. Connect Howly to your portal, connect Howly to Claude Desktop, and you can ask for a health score, a list of broken enrollments, or a full client-ready summary in plain language. For a fix list your whole team can act on, run the AI Audit inside Howly directly and export the ranked findings with one-click links back to HubSpot.

If your portal has grown past the point where any one person can hold its dependencies in their head, this is the fastest way to find out what is actually broken, and Claude never touches the portal to tell you. For the full setup walkthrough, see the [complete guide to Howly for HubSpot](https://blog.howly.io/howly-hubspot-workflow-guide), and if you are choosing between visibility tools, the [comparison of HubSpot workflow audit tools](https://blog.howly.io/best-hubspot-workflow-visualization-and-audit-tools) breaks down how Howly's Claude integration stacks up. [Connect Howly](https://app.howly.io/) to your portal and start with a 7-day free trial. No credit card required.

* * *

*Howly is a read-only HubSpot workflow mapping and audit tool. It maps workflow connections, flags structural issues, and shows the impact of property changes before you make them. Used by RevOps teams and HubSpot agencies managing complex portals at scale.*