AI agents
and 2FA codes
Two-factor sign-in stops a browser agent cold. Giving the agent your authenticator seed fixes that, and it also gives the agent your second factor forever.
The problem with sharing the seed
A TOTP seed makes every future code. If an agent holds it, anything that can read the agent’s context, logs or tools can make your codes too. 2FA was meant to stop exactly that.
How Keyflap handles the code
- The password and the TOTP secret stay in the signer, a separate part of Keyflap, or in Proton Pass or 1Password, which the signer reads only after you approve.
- After you approve on your phone, Keyflap types the 6-digit code into the page itself.
- The browser tells Keyflap the real address of the page. A fake site matches no credential, so it gets no code.
- The agent sees
signed in
ordenied
. It never sees the password, the code or the seed.
Better still, use a passkey. Keyflap can add a passkey to an account and sign in with it. Your other sign-in methods do not change.
High-risk sites always ask
Banks, payments, shopping, email, sign-in providers, code hosting, password managers and government websites always need a new approval with Face ID, a fingerprint or a PIN. For other sites you can set a rule: deny, always ask, ask with Face ID each time, or allow for a set time (24 hours at most) after one approval.
Limits, said up front
- Keyflap is early alpha. Use test accounts first.
- It does not solve CAPTCHAs or hide from bot detection.
- It works with Claude Code, Codex CLI and Hermes Agent (gateway direct messages) today. Other agents need a hook that sends your exact words to Keyflap.