Digital Plan Vault
← Back to AI workflows

AI workflows

Set boundaries before connecting business tools

Check what one connector can read before you use it with real work. Test a harmless lookup and save the account, permissions, and disconnect steps.

Takes
20–30 minutes to review and test one connection
Needs
A Claude account, an app you’re allowed to connect, and a sample record you control
Costs
Depends on the connected app and your Claude plan. Choose a free connector and app if available.

What you’ll make

Check one connector with a sample you control and save the access and disconnect details.

Worked example

A read-only launch brief for a planner bundle

A creator wants a Friday digest from product launch files and keeps invoices in a separate account. They connect an account with access to the shared launch folder and turn on the search tools needed for the test.

The connector finds 'October launch checklist test' in the launch account. A separate fake invoice is inaccessible from that account. The creator keeps a one-page record with the connected identity, review settings, result links, and disconnect steps.

Before you start

  • Pick one use case and an account you are authorized to link. Identify the data the task uses and any account content that would be inappropriate to include.
  • Prepare two sample records with invented information. Make one accessible to the account you plan to connect and keep the other accessible only to a separate account you control.

Name the account and one read you need

Choose one job, one service, and an account you’re authorized to connect. Create a harmless test record with a distinct title. Write down which records you want the account to be able to read.

Example

I want a read-only launch brief. I make a sample called “October launch checklist test” in the account I use for launch files.

That gives me a harmless record I can find independently and a clear reason for the connection.

Read the access request and try one lookup

In Claude, open Customize and look for Connectors. Before approving a connection, confirm the account and compare the access it asks for with the job you chose. If that access suits the job, connect it, then ask in a fresh chat for the sample’s exact title and link. Check any action-approval controls your setup shows.

Example

I check the launch account on the consent screen and ask Claude to find the test checklist by title using read actions.

I expect the matching title and a link. If the consent screen asks for more access than I want to grant, I can pause and work with a permitted export instead.

Save the access record

Open the returned link under the intended account. Save the account, the access requested, any action-approval choices you can confirm, and where to disconnect. Use the provider’s own access view if you need to confirm whether the account can open other records.

Check you can reverse the test

If you connected just for practice, disconnect and check for any remaining grant at the provider. Try the lookup in a fresh chat if you need to confirm the connection is gone. If you’re keeping the connector, save the disconnect steps alongside the sample link.

  • You confirmed the account and the requested access.
  • The test record opens in the intended service.
  • You know where to review or remove the connection.

Optional: Check what the connector can see

If you control two practice accounts, keep one sample in the connected account and one in the other. Try a lookup from the first account and compare the result with the access you expected.

Write down exactly what this connection needs

Start before connecting the service to map the account, read scope, and actions you’ll review.

I want to use [service] in Claude to make [output] from [exact folders, channels, or record types]. I’ll sign in with [intended account or workspace]. The job needs [lookups or actions], and these records are outside the job [excluded sources]. Here’s the consent wording or settings I can see [paste text or add screenshots].

Give me a short access plan with the account to confirm, the provider permissions to inspect, the connector actions to review, and what to do if available access is wider than the job needs. Keep provider access, tool approval, and privacy or data settings in separate rows. Mark anything you can’t establish from what I provided. Give me an allowed and an inaccessible harmless test record to set up, the expected result for each, and where I should record how to disconnect later.

Try an exact-title search with a test record

Run this in separate chats for the allowed and inaccessible samples, then compare with the service itself.

Using only [connector and named workspace if visible], search for the exact test title [sample title] in [location, if appropriate]. Keep this to search or read actions and keep the result in chat. If you find an exact match, return its title, link, and visible location. If there are multiple matches, list the locations so I can identify mine. If you can’t retrieve it, show the query, scope, and what the tool reported. Say whether you saw a permission error, no match, or no available tool. Leave the account’s actual access unconfirmed until I check it in the service.

Troubleshooting

What if Claude requests broad scopes for a tiny task?

Use an account with access only to the intended material if you can grant it that way. Otherwise build the brief from selected excerpts rather than enabling the connector.

What if Claude answers after I disconnect?

Run the test in a new chat with a new sample title it has never seen. Verify the provider's remaining grants and ask for a source link to distinguish retrieval from an unsupported answer.

Downloads

Further reading