Hey there!

If you use the Postman desktop app alongside an AI coding agent like Claude Code, Codex or Cursor, it's worth taking a couple of minutes to check your config. Developers have reported that a Postman update installed its own AI plugin into those agents, along with a command-line tool and a startup hook, without asking first. Postman has since said it will make opt-in clearer.

In this post I'll go through what was reported, what's confirmed and what isn't, what Postman has said, and how to check for the Postman plugin and remove it.

What Postman installed without asking

It started with GitHub issue #14370, opened on 25 October. The developer was running Postman desktop 12.33.4 on macOS. Postman's own log on their machine shows an auto-existing-upgrade event on 24 October, with results saying plugins were installed for Claude Code and Codex. They only noticed because Codex started throwing authentication errors and loading Postman skills they hadn't asked for.

According to the report, that one event:

• Enabled a Postman plugin in Claude Code's ~/.claude/settings.json.
• Added a Postman plugin marketplace and an enabled postman@postman plugin to Codex's ~/.codex/config.toml.
• Installed a Postman CLI at ~/.local/bin/postman.
• Pointed Codex at Postman's remote MCP server, mcp.postman.com, which is where the authentication errors came from.

MCP (Model Context Protocol) is the standard way coding agents connect to outside tools and data, so the agent now had a Postman connection it could reach for.

Other users in the thread reported the same thing. One updated from 12.31.3 to 12.35.1 and found the CLI and the Claude Code and Codex plugins, with a Cursor plugin already present. Another found Postman entries in GitHub Copilot's settings that were still there after using Postman's own uninstall option.

Why the startup hook bothered people

The part that drew the most criticism is a session-start hook, a small instruction the agent picks up every time it starts. Postman's hook tells the agent to load its postman:api-engineer skill before most API work, from designing and mocking an API to testing, documenting and deploying it. In practice, your agent gets nudged towards Postman's tools whenever you touch an API, whether or not that project uses Postman.

Codex normally asks you to approve a hook like this before it runs. On the original reporter's machine, though, the Codex config already held a "trusted" record for Postman's hook, and the reporter pointed to code in Postman's installer that approves its own hook. Postman's docs for its command-line installer say the same thing: the installer approves the hook and only leaves it for you to review if that fails.

For the desktop app, this isn't settled. The second reporter found no trust record on their machine, and a commenter testing Codex 0.162.1 said a hook without one is simply skipped. Postman's desktop app page now says the hook is untrusted by default and Codex will ask you to review it. So the hook may not have run everywhere the plugin landed.

What hasn't been shown

Some of the coverage made this sound like Postman was quietly collecting API data. The evidence doesn't support that. The original reporter says the report doesn't claim to establish data exfiltration, and nobody in the thread has shown data leaving a machine. What is documented is an app changing other tools' configuration without consent, which is a real problem on its own.

The detailed reports are also from macOS. I couldn't find a documented Windows or Linux case, though that isn't the same as those machines being unaffected.

What Postman has said

Postman's head of product, Sarabmeet Kallar, told RuntimeWire on 3 November that the company will make opt-in for its coding agent plugins clearer, particularly for existing users. According to RuntimeWire, Postman is also reviewing the hook approval and session-start behaviour flagged in the report.

Postman's documentation now says the desktop app doesn't install the plugin automatically and only does so when you pick the menu option. It adds that if the app installed the plugin automatically before, you can remove it. That's the closest thing to an acknowledgement so far, though it doesn't say which versions did the automatic install or when it stopped.

How to check if Postman installed plugins on your machine

If you've had Postman desktop installed and signed in recently, look for these:

• Claude Code: postman@claude-plugins-official in ~/.claude/settings.json, or a Postman plugin in Claude Code's /plugin menu.
• Codex: a [marketplaces.postman] table, a [plugins."postman@postman"] table or a hooks.state entry mentioning Postman in ~/.codex/config.toml.
• Cursor: a postman folder in ~/.cursor/plugins/local, or a Postman plugin under Cursor Settings > Plugins.
• Copilot: Postman entries in ~/.copilot/settings.json.
• The CLI: a postman file in ~/.local/bin.
• On a Mac: search Postman's log at ~/Library/Application Support/Postman/logs/main.log for auto-existing-upgrade.

If none of those are there, you're fine. If they are and you installed them on purpose, there's nothing to do.

How to remove the Postman plugin

Postman's desktop app docs give the official route:

1. In Postman, go to Help > Postman CLI and Plugins > Uninstall Postman Coding Agent Plugin, and click Uninstall. This covers Claude Code, Codex and the local Cursor copy.
2. Restart your agents so they reload their config.
3. If Cursor has a marketplace copy, remove it under Cursor Settings > Plugins.
4. For any other agent, run npx @postman/postman-plugin remove. Add --dry-run to preview what it will do, or --agent copilot to target a single agent.

Then go back through the checklist above. The Copilot report suggests the menu option doesn't catch everything, and the original reporter argued that uninstalling should also undo trusted hooks, MCP connections and the CLI. If a Postman entry is still in a config file, back the file up and delete the entry by hand. If the CLI is still in ~/.local/bin and you never installed it, you can delete that too.

The takeaway

Coding agents read their plugins, hooks and MCP servers from plain config files in your home folder, and any app running as you can write to those files. Postman didn't need special access to do any of this, just the permissions every app you install already has. That means the agent you rely on is only as predictable as the least careful tool on your machine.

So it's worth checking those config files now and then, the same way you'd review your browser extensions. If you use Postman, look for the entries above and remove anything you didn't choose. And if you build tools that plug into coding agents, ask before you install anything.