Anthropic today rolled out a new policy that kicks automated script tasks out of the regular subscription pool, slapping them with a hard $20 per month usage cap. But the company explicitly said that "interactive typing chats" from a real person at the terminal remain unlimited.
Within hours, the developer community fired back. An open-source project called claude-p disguises scripted API calls as human typing, successfully sidestepping the paywall.
Here's how it works: first, it silently launches a pseudo-terminal (PTY) in the background and opens a legitimate Claude Code interface. Then it uses code to mimic a person — typing prompts on the keyboard and hitting enter. After Claude spits out its answer, the tool grabs the output and passes it to an external program.
Thanks to this camouflage, Anthropic's servers see all those high-frequency machine requests as a real human typing chat sessions. Third-party tools can therefore keep using the original subscription quota, unconstrained by the $20 cap.
The project's README snarks: "Trying to control how people use your product via the client side is a complete waste of effort."