Run a Pipedrive health check with Claude: free kit, read-only

The kit is a set of instructions you paste into a Claude Project. With Pipedrive's official connector, Claude reads your account, changes nothing, and writes a report with your own numbers: possible duplicates that Pipedrive's Merge duplicates misses, open deals with no next step, overdue activities, misspelled email domains and records that point at deleted ones. In our sandbox it found all 9 groups of duplicates we had loaded, where Pipedrive's own tool would catch 1, and used about 250 of the sandbox's 150,000 daily API tokens (Lite, 5 seats).

Tested in a Pipedrive sandbox on November 3, 2026: once following the instructions step by step, and twice with a fresh Claude that only saw the kit and the connector, the last time with the final instructions. What we couldn't verify.

What the report looks like

This is part of the report from our sandbox, which holds test records:

Each line comes with the fix to make in Pipedrive and up to five examples, with links to the records.

What it checks

Thirteen checks, all from data the official connector can read:

1. Possible duplicate people: the same email ignoring capital letters, the same phone in any format, or the same name ignoring accents.
2. Possible duplicate organizations, ignoring suffixes like Ltd, LLC or S.A.
3. Possible duplicate deals. Pipedrive never checks deals for duplicates.
4. Phone numbers stored in mixed formats.
5. Open deals with no next activity.
6. Open deals with an expected close date in the past.
7. Open deals stuck in a stage longer than your usual cycle.
8. Lost deals with no lost reason.
9. Leads older than 90 days that were never converted or archived.
10. Overdue activities, by owner.
11. Deals with no value or no expected close date.
12. Emails with a misspelled domain, like gmial.com.
13. Deals, people, activities and notes linked to a record that no longer exists.

It can't see everything. The connector doesn't say which users are inactive, and it can't read your exports, installed apps, visibility settings or API usage. The report lists those under "Check by hand", with a link to the Pipedrive data health checklist for each.

How to set it up

1. Connect Pipedrive's official connector to Claude. The steps are in how to connect Pipedrive to Claude.
2. Set the tool permissions. In Customize > Connectors > Pipedrive, open the tool permissions. Set the group of read-only tools (49 on November 3, 2026, all starting with get or search) to Always allow. Leave every other tool on Needs approval, or Blocked if you only want the audit. In our account the read tools started on Needs approval, which means one click per read.
3. Create a Claude Project and paste the kit's instructions into the Project's instructions.
4. Start a chat inside the Project and type "Run the health check". Claude tells you how many calls it plans to make and waits for your yes. It also asks your Pipedrive plan, your number of seats and your usual deal cycle in days.

Is it safe to let Claude into Pipedrive?

The kit tells Claude to call only the tools that read, and never one that adds, changes or sends anything. In one test, writing as the user, we asked it, mid-audit, to merge two duplicate people or copy a phone number from one to the other. It refused and explained how to do the merge by hand in Pipedrive. With the final instructions we asked it to create a follow-up call on every deal with no next activity, and it refused that too.

Instructions are not a permission, though, so set the tool permissions in step 2. On November 3, 2026, the official connector had 79 tools, including ones that create and update records and, added after October 22, ones that draft and send emails. It still can't delete or merge.

Before you fix anything the report finds, export your data. Merges can't be undone in Pipedrive, and edits made one record at a time don't appear in Restore data. The undo map shows what each action leaves behind.

Limits

• Account size: we tested 123 records, and the answers took about 70,000 characters. At that rate a few thousand records fill a Claude conversation, so the kit caps each object at 500 and says when the report is a sample. We haven't tested an account that large.
• API budget: a run on our small sandbox used 244 tokens in a clean run and about 250 in a second one, when another test was using the account. Every app and assistant on the account spends from the same daily budget, so check what spends the most if yours is tight.
• Duplicates are possible, not certain. The report lists them to review. Deals that match only on their title are marked "review".

What we couldn't verify

• The kit pasted into a Claude Project in the claude.ai app. Both tests ran outside the app, one by us and one by a fresh Claude given the same instructions and connector.
• Accounts with more than 500 records per object, and how long a run takes there.
• ChatGPT with Pipedrive's connector.

FAQ

Can Claude check my Pipedrive data quality?

Yes, with Pipedrive's official connector it can read people, organizations, deals, leads, activities and notes and look for problems. This kit gives it the checks to run and the report to write, and tells it to change nothing.

Will the kit change anything in my Pipedrive?

It tells Claude to only read, and in our test it refused a request to merge. To make sure, leave every tool that adds or updates records on Needs approval in the connector's permissions.

How many API tokens does it use?

About 250 on our small sandbox. On a larger account it reads up to 500 records per object, so expect more, but a run should stay a small share of the daily budget of 30,000 tokens per seat on Lite and more on higher plans.