Postman app altered developers’ AI coding tool settings

Postman’s macOS client is reported to have silently modified developers’ local AI coding environments by installing background extensions and registering new plugin repositories for internal tools in Codex and Claude Code. The changes occurred without explicit user authorisation and included persistent session startup hooks designed to load Postman tooling as default entry points for API development tasks.

In practice, this meant that when a developer began a session, the AI models could prioritise Postman capabilities over other tooling, effectively routing API design, testing, and deployment queries through Postman’s own skills. The issue came to light after a misconfigured remote MCP endpoint repeatedly failed authentication, which helped reveal the hidden configuration changes in user directories.

OpenAI Codex and Claude Code were both affected, with the Postman extension and standalone CLI being activated without browser prompts or OS permissions being shown.

Postman has defended the bundled delivery as an intentional convenience feature aimed at simplifying onboarding, while critics describe the behaviour as workflow hijacking and unauthorised modification of independent developer configurations. In response, the vendor has promised explicit authorisation prompts in future releases.

In the meantime, affected developers are advised to inspect local configuration files for Codex and Claude Code, purge Postman entries from plugin lists, and manually remove any startup hooks that were installed. The article attributes the concerns to engineers and references a Postman app support GitHub issue documenting unauthorised local file modifications.