Create, test, and troubleshoot a custom tool

Prerequisites

You need AI Tools management access and an endpoint implemented by your team or provider. Agree the request and response contract with its owner. Review the remaining tool allowance and the name, description, parameter, and response limits shown for your plan; do not assume another plan has the same limits. Use test credentials and records for operations that change external data.

Configure the tool

1. Open AI Tools → Add Tool. Describe exactly when the agent should use it and what information it needs.
2. Add parameters with the supported type: string, number, boolean, date, or time. Mark required parameters and write clear descriptions.
3. Add validation offered by the editor, such as numeric range, pattern, or not-past constraints. Mark system-provided values only when they are actually supplied by the system.
4. If a parameter obtains valid values from another tool, choose the parameter's tool dependency. Test that the prerequisite tool returns the intended options.
5. Select the supported HTTP method and enter the endpoint URL and headers. Link the intended integration and expected events if required by your contract.
6. Save and run the tool test. Inspect reachability, HTTP status, response time, and error text.
7. Enable the tool and ask the AI to perform the operation with a controlled customer record. Verify the actual result in the remote system.

Maintenance

Edit a custom tool when its endpoint or contract changes. Disable it while correcting a fault. Read the deletion confirmation before removing it; inspect linked integrations and workflows that may depend on it.

Troubleshooting

An unreachable endpoint needs network/URL diagnosis. A rejected request needs credential, required-field, or contract checks. A successful HTTP status does not prove the business result. Inspect the provider record and prevent duplicate side effects before retrying. Never publish real endpoint secrets in screenshots or documentation examples.