Put WorkerKit to work with your dot.

Before you start

What you need.

For discovery: browse the public kit directory at workerkit.ai/kits. Public kit reads need no WorkerKit account.

For Fleet: a WorkerKit account and an account admin to approve OAuth access. You can create an account during approval.

For execution: a worker, the app connections its job needs, an enabled deployment, usable model funding and budget headroom.

For email reports: a sending mailbox connected in WorkerKit and a receiving mailbox connected to your dot.

For custom connections, follow OpenAI's connection guide: enable developer mode in Settings under Security and login where your workspace permits it, then add the server from Plugins. Use the dot's app controls to enable the connection for that dot and review its advertised tools.

Start with the WorkerKit MCP endpoint above for fleet operations and kit discovery. Add Knowledge separately when you need published worker reports.

The case for connecting at all is on the OpenAI Dots page.

Account access

Connect WorkerKit.

Use OAuth to connect your account and approve the operations the job needs. Fleet includes kit discovery.

1. 01

   Add WorkerKit and approve access

   Create the WorkerKit connection in ChatGPT Plugins. Sign in through OAuth and have an account admin approve the account, permissions and expiry.

2. 02

   Read the account and scope

   Enable Fleet for the dot and inspect its advertised tools. Call key_info, onboarding_get and workers_list. Confirm the intended account and the scopes needed for the requested work.

3. 03

   Choose a worker

   Use an existing worker or search the included catalog. For a new kit, use kit_install_preview to review setup and permissions, then kit_install within the approved task. Request deployment with deploy: true when manageDeployments is granted, or deploy the worker in WorkerKit.

• Custom MCP connection Name: WorkerKit Connection: https://mcp.workerkit.ai/workers Authentication: OAuth

The agent operating reference carries the tool workflow. Continue with your first result after the account connection succeeds.

Fleet scopes govern management and result access. Runs use the app grants assigned to the selected worker; receipts can contain information from those apps. Review both layers for the job.

Human and agent

Get your first useful result.

Start with a small job, inspect its result and then make it recurring.

1. 01

   Review the job and access

   Read the instruction and worker permissions. Agree on the input, output and any app changes the first run should make. For discovery or research, use read access; for action kits, review the specific writes.

2. 02

   Check deployment and funding

   Confirm the worker and deployment are enabled, a model is selected, funding is usable and per-run, daily and account budgets have headroom. Use onboarding_get and account_usage to identify the next setup action. The person completes any app sign-in or wallet checkout in the trusted page.

3. 03

   Run once and inspect

   Run the approved job and keep its run ID. Read the receipt for the output, status and costAuthority, then the settled cost when available. Compare the result with the request and report any partial work. Check an existing run after a timeout before starting another.

4. 04

   Choose the ongoing cadence

   Configure the report destination and confirm receipt of the test report. Once the result is useful, enable the approved schedule with its time zone. Review the first scheduled run.

Use WorkerKit to pause a worker, cancel a running job or manage its schedule. Manage the dot's Fleet connection separately in ChatGPT. Each control has its own purpose.

For a Skipped run, read skipReason and complete the named readiness or funding step before another attempt. Report connection success and job completion separately.

Report delivery

Send reports to the inbox your dot reads.

Use your connected mailbox to bring worker results into the dot's conversation.

1. 01

   Connect the receiving mailbox to your dot

   In the dot's plugin controls, connect the personal mailbox that will receive reports. Ask the dot to find a known message to confirm access.

2. 02

   Connect the sending mailbox in WorkerKit

   Under WorkerKit Connected apps, connect a Google or Microsoft mailbox for sending notifications. This sending account can differ from the receiving mailbox connected to your dot.

3. 03

   Choose the report recipient

   On the worker, open Notifications, add an Email notification and enter the receiving address from the mailbox step above. Choose the delivery condition and what the report should include, then save.

4. 04

   Run and review the report

   Complete the first-run readiness checks, run the worker and confirm its delivery status. Find that report in the receiving mailbox by sender and time. Ask the dot to summarize the work and link to the message. Use the worker and run links to inspect details or continue a run waiting for input.

• Ask your dot Find the latest WorkerKit report in my connected email. Summarize the completed work and anything that needs my decision. Include the original message and any worker or run links in the report.

Open Connected apps to configure the sending account, or open your workers to manage runs and schedules.

Operating rules

From discovery to useful work.

Choose the route, review the job and bring back the result.

Quick answers

Symptoms, causes, fixes.

Use the returned account, run and delivery states to choose the next action.