Your AI assistant needs to work in WordPress: draft posts, review content, update pages. To do that through the REST API, it needs credentials.
The built-in option, Application Passwords, was designed for humans pasting a username and password pair into an app once. AI agents don’t work that way:
- Agent credential vaults store one secret, not two. Most agent platforms (including Muse’s custom connectors) have a single API-key field. An Application Password is a username plus a password — it simply does not fit, and the connection gets refused.
- Application Passwords rely on Basic Auth. Many shared and managed hosts strip or block the Authorization header, so perfectly valid credentials fail with no useful error.
- One leaked password pair is hard to scope. You end up handing an agent credentials tied to a real user account, and revoking means digging through profile pages.
Keypaw takes a different approach: one token, built for agents.
- Generate a token in Settings Keypaw. It is shown once; only its SHA-256 hash is stored.
- Map the token to a dedicated WordPress user you choose (for example, an Editor account created just for your agent). The token can do exactly what that user can do — no more.
- Your agent sends the token in a simple header (
X-Agent-Token) over HTTPS. No Basic Auth, no username, nothing else to configure. - Rotate or revoke the token in one click whenever you want — the old token stops working instantly.
- Agent self-discovery: the plugin publishes a small public manifest (
/wp-json/keypaw/v1/manifest) describing how to connect — header name, site URL, and the mapped account’s capabilities. Point your agent at your site and it can figure out the rest. - Copy-paste setup: when you generate a token, the plugin gives you a ready-made setup snippet. Paste it to your agent and it is connected.
Security by design: HTTPS only, hashed storage, constant-time comparison, no telemetry, no external calls, and nothing is authenticated when no token is configured. The plugin adds no public write endpoints of its own — it only authenticates standard REST API requests as the user you mapped.
Born in production: Keypaw was written because an AI assistant reviewing drafts on a real site could not use Application Passwords at all. It has run that workflow daily since.
Note: if your tooling can use Application Passwords (local scripts, your own server code), they remain a fine built-in option. Keypaw exists for the growing number of agent platforms that cannot.