PRACTICAL FIELD GUIDE

Revoking a key and verifying that access stops

Key revocation should be followed by a test through the actual data path. The current key-resolution cache can make a recent revocation take a short time to be observed by an already warm worker.

MCPBackend editorial team · · Examples are illustrative

Put it into practice

  1. Identify the exact key and replace any legitimate integration dependency first.
  2. Revoke through the supported management path and avoid logging the secret during testing.
  3. Verify that new requests stop succeeding after the applicable cache window and investigate continued access.

What this looks like

ILLUSTRATIVE EXAMPLE

A recently used key remains resolved briefly in a worker's short-lived cache. The test needs timing context rather than an immediate blanket conclusion.

A boundary to keep clear

Do not describe revocation as an instantaneous global kill switch without checking the implementation and observed result.

MCPBackend context

MCPBackend row policies apply to anonymous callers and end-user tokens. Machine API keys bypass those row policies and are instead constrained by their own table permissions. Owner mode is based on the calling user's identity; it is not an organization-membership or arbitrary custom-policy engine.

Take the next step

Inspect the current project contract, try the change with disposable data, and verify the result through the same credentials your app will use. Record the expected response and one denied-access case before release.

Sources and further reading

These references explain the underlying protocols and design principles. For supported MCPBackend operations and exact request shapes, inspect your project’s generated API contract.