ANTHROPIC PLUGIN PORTAL — OPENED SEPT 25, 2026

Cybersecurity News: Are Claude Code plugins safe? The portal opened with auto-update ON by default.

On September 25, 2026, Anthropic opened the Claude Code plugin portal — the official channel for plugins that extend your coding agent with commands, skills, and bundled MCP servers. One default matters more than everything else: auto-update is ON, meaning the code inside your agent's toolbox can change overnight without asking you. Official plugins get more review than sideloaded ones, but every plugin is a supply-chain surface — whoever controls the update controls the code your agent runs, with your file and network access. This free 60-second quiz audits your plugin provenance: official vs sideloaded sources, auto-update posture, and the bundled MCP servers most people never check.

Key figure: Reported cybercrime losses in the United States reached $16.6 billion in 2024, up 33% from 2023 (FBI IC3, 2024 Internet Crime Report).

In short: Claude Code plugins range from well-reviewed official ones to unreviewed sideloads, and since the plugin portal opened on September 25, 2026, auto-update has defaulted to ON — so plugin code can change under you silently. Plugins are only as safe as their provenance, their update posture, and every MCP server they bundle. This 6-question quiz tells you whether you know what's actually inside your agent's toolbox.
Who Anthropic — plugin authors and anyone running Claude Code with third-party plugins
What The official Claude Code plugin portal launched with auto-update defaults ON — plugins (commands, skills, bundled MCP servers) can change without prompting
When September 25, 2026
Where Claude Code's plugin portal; plugins install from the portal or sideloaded from other sources
Sources Anthropic's plugin portal launch (Sept 25, 2026); portal defaults and plugin architecture as documented

The 60-second plugin provenance quiz

Answer honestly. Here, "Yes" is the safe answer.

Answer the 6 questions above to get your drift-risk verdict.

This self-assessment is informational. It reflects the plugin portal's documented launch and defaults plus standard supply-chain practice — it is not legal advice.

Why auto-update defaults change the threat model

What this check does NOT do

Open the plugin-provenance-verifier on GitHub →

Open-source tooling: plugin inventory, source verification, auto-update pinning helpers, and bundled MCP server enumeration for Claude Code.

FROM INGENIOUS AI
Rogue-Agent Defense watches what your agents and their plugins actually do — permissions, tool calls, and anomalies — in real time.
All 11 platforms: DisclosureGuard  ·  ModelMinder  ·  TermsWatch  ·  Vibe-Code Audit  ·  Rogue-Agent Defense  ·  BreachHunter  ·  ChatBreaker  ·  Agent Flight Recorder  ·  Agent Auth Gateway  ·  DomainGuard  ·  ContentCraft AI
Our platforms are being upgraded and return November 1 — get early access now:
Get early access on WhatsApp →

What is the Claude Code plugin portal?

The plugin portal is Anthropic's distribution channel for Claude Code plugins, opened on September 25, 2026. It lets developers publish and users install plugins — bundles of commands, skills, and MCP servers that extend Claude Code. Plugins can be installed from the official portal or sideloaded from other sources, and auto-updates default to ON.

Share this check: XLinkedInReddit

Cybersecurity news: frequently asked questions

Are Claude Code plugins safe?

Mostly — but "plugin" is not a single trust level. Official plugins from Anthropic's portal get more review than sideloaded ones, and since September 25, 2026 the plugin portal has shipped with auto-update defaults ON, meaning plugin code can change under you without a prompt. Safety depends on provenance (official vs sideloaded), whether updates are pinned or reviewed, what bundled MCP servers a plugin includes, and what permissions the plugin's tools hold.

What is the Claude Code plugin portal?

The plugin portal is Anthropic's distribution channel for Claude Code plugins, opened on September 25, 2026. It lets developers publish and users install plugins — bundles of commands, skills, and MCP servers that extend Claude Code. Plugins can be installed from the official portal or sideloaded from other sources, and auto-updates default to ON.

Can Claude Code plugins auto-update?

Yes. Since the plugin portal opened on September 25, 2026, auto-update has been the default — plugin code, commands, and bundled MCP servers can be replaced with newer versions without asking you first. That is convenient, and it is also a supply-chain attack surface: whoever controls the update controls the code your agent runs.

How do I audit Claude Code plugins?

List every installed plugin and its source (official portal vs sideloaded); check whether auto-update is on or pinned; inspect each plugin's declared tools and permissions; enumerate any MCP servers bundled inside plugins; and review update history — would you notice if a plugin's code changed last night? Re-audit whenever a plugin updates or you install a new one.

What is plugin drift risk?

Plugin drift is the gap between the plugin you installed and the plugin running today. With auto-updates on, code, prompts, and bundled MCP servers can change between sessions — including changes pushed after the maintainer's account was compromised. Drift risk is highest for sideloaded plugins with auto-update enabled and no integrity verification.

Do Claude Code plugins include MCP servers?

They can. Plugins are bundles that may include MCP (Model Context Protocol) servers giving the agent tools over your files, network, and apps. A plugin's trustworthiness is therefore the combined trust of its own code plus every bundled server — each server is an independent attack surface with its own permissions.

More from Ingenious AI

Related checks — Cybersecurity News

Filed under Cybersecurity News by Ingenious AI — see the related checks above for more in this series.