Postman silently auto-installs AI plugins and attempts to share all API data

A macOS user reports that opening Postman Desktop 12.33.4 automatically installed integrations for Codex and Claude Code, along with a Postman CLI, and changed global configuration in both coding agents. Postman’s log recorded a successful automatic installation for existing users. In Codex, the installation added a remote Postman MCP connection that failed authentication and a startup hook directing the agent to load Postman’s API-engineering skill for a broad range of API tasks. The user also found that the installer’s code attempts to approve Postman’s own hook by writing its trust record. The report distinguishes this code behavior from proof of who created the observed trust record, and explicitly does not claim that data was exfiltrated.

The user frames the changes as a serious consent and security-boundary failure: opening an API client should not alter unrelated tools or make Postman their default workflow. The sequence is based on one Mac and has not been reproduced on a clean installation. The report asks Postman to halt automatic installations for existing users, remove installer-side hook approval, disclose the feature flag and consent flow behind the behavior, and provide a complete rollback that prevents removed integrations from returning. It also calls for narrowing the startup instruction to cases where users explicitly request Postman. The user removed Postman and its integrations after preserving evidence; the issue remains open, with no assignee or labels and no maintainer response shown.