The Postman AI plugin controversy began when Postman Desktop 12.33.4 automatically installed plugins for OpenAI Codex and Claude Code on existing installs, without a clear opt-in, and tried to approve its own startup hook. Postman said on November 5, 2026 that it is reverting the change. If you use Postman, remove the plugins, check your agent configs and rotate any API keys you can’t account for.
Developers treat their API client like a toolbox: boring and trusted. So when a GitHub issue titled “[SEVERE] Desktop auto-installs Codex/Claude plugins; installer attempts to trust its own startup hook” landed in late October, it hit a nerve (GitHub issue #14370). It shot to the top of Hacker News under the headline “Postman silently auto-installs AI plugins and attempts to share all API data,” and the comments turned into a referendum on whether to keep using Postman at all.
Key takeaways
- Postman 12.33.4 added Codex and Claude Code plugins plus a Postman CLI on existing installs, logged as “auto-existing-upgrade.”
- The issue’s author said it does not prove data exfiltration. The bigger problem is consent and an installer trusting its own hook.
- Postman staff said the change is being reverted and the experience will be redesigned with more control.
- Your API keys are at risk mostly from what Postman already stores and syncs, not just the plugin.
- Git-based clients like Bruno keep requests in local files you control.
What did the Postman AI plugin actually do?
According to the issue, opened October 25 by GitHub user indicava, the desktop app:
- Installed a Codex plugin (v1.1.0), a Claude Code plugin (v2.0.1) and a Postman CLI in
~/.local/bin. - Edited the global Codex config at
~/.codex/config.tomland Claude settings and plugin records. - Tried to mark its own SessionStart hook as trusted in the Codex config. That hook told the agent to load a Postman API-engineering skill for broad API tasks, not only when you asked for Postman.
- Enabled a remote Postman MCP server, which failed to authenticate on the author’s machine.
A coding agent plugin is an add-on that gives tools like Codex or Claude Code extra instructions, commands or connections to outside services. When one loads automatically at the start of every session, it can steer what the agent does and which servers it talks to.
The original report says plainly that it “does not claim to establish data exfiltration.” The Hacker News headline’s “share all API data” goes further than the evidence in the issue. The core complaint is that Postman changed other tools’ trust settings without asking.
How did Postman respond?
On November 5, a Postman member, preethammavin, wrote in the thread: “In the meantime, we are reverting this change,” adding that the company “will be re-designing the experience with the right level of control.” Another commenter, yumeria, said “we bundle the Postman CLI and the agent plugins with the desktop app” and that the team would “make the opt-in clearer, especially for existing users.” The thread was then closed and locked. Users who had already been affected asked what happens to their machines, and that question didn’t get a direct answer.
Why the Postman AI plugin matters for your API keys
The plugin is the headline, but the deeper worry raised on Hacker News is what Postman already holds. Collections, environments and request history often include auth headers and tokens, and with a signed-in account much of that syncs to Postman’s cloud. Commenters argued that history can keep the resolved values of templated secrets, so a {{api_key}} variable doesn’t guarantee the real key isn’t stored. We couldn’t confirm that behavior in Postman’s documentation for every version, so treat it as a reason to check, not a settled fact.
Others were blunter about the product’s direction. User pants2 wrote that “It started as an excellent little developer tool for testing APIs” before arguing the company now leans on forced logins, data collection and AI features (Hacker News).
How do you audit Postman and rotate your API keys?
- Update or roll back Postman to a version without the auto-install once Postman ships the revert, and check its changelog.
- Remove the plugins. In Codex, open
~/.codex/config.tomland delete Postman plugin, hook and MCP entries. In Claude Code, run/pluginand check~/.claude/settings.jsonfor Postman hooks or MCP servers. - Delete the CLI from
~/.local/binif you didn’t install it yourself. - Clear history in Postman and review environments for real secret values in the “initial value” field, which is the part that syncs.
- List every key in your workspaces. Payment, cloud, database and third-party API tokens first.
- Rotate anything sensitive. Create a new key, deploy it, then revoke the old one. Check the provider’s audit logs for use you don’t recognize.
- Move secrets out of the client. Pull them from a secrets manager or local environment variables at runtime, and use short-lived, narrowly scoped tokens.
- Tell your security team if this happened on a work machine. Changed agent configs are worth a look.
If a key does leak, our checklist on what to do after a data breach covers the follow-up. Storing those keys in a vault instead of a client is easier with a password manager; see our take on whether password managers are safe.
What are the best Postman alternatives?
Commenter dham summed up the appeal of one option: “We went to bruno years ago and has been good. Nice to be able to store the request definitions in git” (Hacker News). Here’s how the main choices compare:
| Tool | Where requests live | Account required? | Best for |
|---|---|---|---|
| Bruno | Plain-text files in your repo | No | Teams that want requests reviewed in git |
| Insomnia | Local, git or cloud sync options | Optional for local use | Postman-like interface with more storage choices |
| HTTPie | Terminal commands, plus a desktop app | No for the CLI | Quick tests and scripts |
| curl | Shell scripts | No | Automation and zero dependencies |
| VS Code REST Client / JetBrains HTTP Client | .http files in your project | No | Developers who live in their editor |
Check each tool’s current terms before you switch. Desktop apps change policies too, so the safest pattern is any client where secrets come from environment variables and nothing syncs by default.
Our take: what this means for you
Plenty of developers want an agent that understands their APIs, so bundling AI features is fine. Quietly rewriting another tool’s config and pre-approving your own hook is what security teams flag as malware, whatever the intent. Postman reversing course was the right call, but the episode shows how much access a “harmless” desktop tool has.
Our advice: don’t keep long-lived production keys in any API client. Audit your coding agent configs regularly, since they’re a growing attack surface. Our piece on what agentic AI is explains why. If you’re on a team, moving to git-stored requests like Bruno makes changes reviewable. With git-stored requests, a change like this would have shown up in review on day one.
Frequently asked questions
Did Postman steal my API keys?
There’s no public evidence of that. The GitHub report specifically says it doesn’t establish data exfiltration. The concern is unconsented changes to AI agent configs, plus how much sensitive data API clients store in general.
How do I check if the Postman plugin is installed?
Look in ~/.codex/config.toml for Postman plugin, hook or MCP entries, and check Claude Code’s plugin list and ~/.claude/settings.json. A postman binary in ~/.local/bin you didn’t install is another sign.
Is Postman still safe to use?
It’s a widely used tool, and Postman says it’s reverting the change. Whether you trust it depends on your risk tolerance. Either way, avoid storing production secrets in synced workspaces.
Is Bruno a good Postman alternative?
For many teams, yes. Bruno stores collections as plain files you can commit to git and doesn’t require a cloud account, which keeps requests reviewable and local.
Should I rotate my API keys?
If keys were stored in Postman, rotating the sensitive ones is cheap insurance. Start with payment, cloud and database credentials, and check provider logs for unexpected activity.

Community Insights
0 Comments