What Are AI Plugins? Plain-English Glossary Definition (2026)

What it is: A plain-English definition of AI plugins — what they are, how they differ from skills and integrations, and how they work in Claude, ChatGPT, and other major AI platforms in 2026.
Who it is for: Anyone reading about AI tools and seeing “plugin” without a clear definition.
Best if: You want a foundational definition that holds across platforms.
Skip if: You already understand the plugin pattern from another platform context.

The short definition

An AI plugin is a packaged extension that adds new capabilities to an AI assistant. Plugins let the AI do something it could not do natively — access a specific tool, query a specific data source, perform a specific kind of action — without the user writing code.

If you have ever installed a plugin in a web browser (ad blockers, password managers, grammar checkers) or in WordPress (Yoast SEO, contact forms, page builders), you already understand the pattern. AI plugins extend AI assistants the same way: they add specific capabilities without you needing to know how the underlying code works.

What plugins typically do

• Connect to external services. A QuickBooks plugin lets the AI read your books. A Slack plugin lets it post to your team. A Salesforce plugin lets it update records.
• Add new commands or workflows. A research plugin might add a /research command. A summarization plugin might add a /tldr workflow.
• Embed domain expertise. A legal-review plugin packages legal-document analysis patterns. A code-security plugin packages security-audit logic.
• Bridge to other tools. A browser-control plugin lets the AI navigate websites. A file-system plugin lets it read and write local files.
• Add UI surfaces. Some plugins add new screens, dashboards, or interaction patterns to the AI assistant.

Plugins vs Skills vs Integrations: the 2026 vocabulary

These three terms are often used interchangeably, but in 2026 they have distinct meanings on most major AI platforms:

• Plugins are packaged extensions, usually distributed through a marketplace, that add capabilities. They are the broadest category.
• Skills are reusable instruction bundles that customize AI behavior for specific repeated tasks. They are more like “saved prompts plus rules” than full plugins.
• Integrations are connections to specific external systems (QuickBooks, Slack, HubSpot). Native integrations are typically built by the AI platform; plugins often wrap them.

Practically: when Anthropic mentions Claude for Small Business plugins, they mean the Claude Code plugins that extend the Claude assistant with new commands, agents, and capabilities. When ChatGPT mentions plugins, they mean the GPT Store offerings that add capabilities to ChatGPT. The word is consistent across platforms even if implementations differ.

How plugins work in Claude

In the Claude ecosystem, plugins are commonly distributed for Claude Code. A Claude Code plugin can include:

• Commands — new slash commands you can run.
• Agents — specialist sub-agents that Claude can spawn for specific tasks.
• Skills — reusable instruction bundles included in the plugin.
• Hooks — scripts that fire on Claude lifecycle events (pre-commit, post-response, etc.).
• MCP servers — Model Context Protocol servers that expose new tools and data sources.

Once installed, plugins make their capabilities available across your Claude Code sessions. You do not need to install them per-project; they are typically global to your Claude Code install.

How plugins work in ChatGPT

In ChatGPT, plugins are commonly distributed through the GPT Store as Custom GPTs. A Custom GPT can include:

• Custom instructions — the personality and constraints for that GPT.
• Knowledge files — documents the GPT can reference.
• Actions — API calls the GPT can make to external services.
• Code Interpreter access — sandboxed Python execution.

ChatGPT users can install Custom GPTs from the public GPT Store or build their own private ones.

How plugins work in other AI platforms

• Gemini has Gems (Custom GPT equivalent) and Extensions (connecting to Google Workspace, Maps, etc.).
• Perplexity has Collections and Focus Modes, plus Spaces that function as plugin-like containers for ongoing topics.
• Mistral, Cohere, Microsoft Copilot all have plugin or extension systems with similar concepts.

Why plugins matter

• Capability extension without code. Most users cannot write code to extend their AI. Plugins remove that barrier.
• Reusable across sessions. Install once, use forever. Your AI accumulates capability over time.
• Community contribution. The plugin ecosystem allows specialists to package their expertise for others.
• Vendor differentiation. AI platforms increasingly compete on plugin marketplace depth.
• Foundation for agentic AI. Plugins are how agents get the tools they need to act on the world.

Common plugin pitfalls

• Trust and security. A plugin runs with your AI permissions. Vet what you install. Especially watch plugins that access financial, customer, or healthcare data.
• Quality variance. Plugin marketplaces have wide quality range. The top 10 percent of plugins are excellent; the bottom 50 percent are abandonware. Check ratings and recent updates.
• Privacy implications. Plugins may send data to external services. Review what data flows where before installing.
• Plugin sprawl. Installing too many plugins clutters your AI experience. Periodically audit and remove what you do not use.
• Vendor lock-in. Plugins built for one platform usually do not work on another. Plan your platform commitment with this in mind.

Frequently asked questions

Are plugins the same as MCP servers?

Related but not identical. MCP (Model Context Protocol) is an open standard for connecting AI to external tools and data. A plugin can include MCP servers as part of what it provides, but plugins are typically broader (they may include commands, agents, hooks, knowledge files in addition to MCP servers).

Do plugins cost extra beyond my AI subscription?

Depends on the plugin. Many are free. Some charge a separate subscription. Some are free but consume more API tokens, which costs more in usage-based pricing.

Can I build my own plugin?

Yes. Most platforms publish plugin-development SDKs. Building requires light coding for full plugins; building Skills (a subset of plugin capability) can often be done with no coding.

Are plugins safe?

The plugin itself runs in the AI platform sandbox, but it can take actions on your behalf in connected services. Treat plugin permissions like you would treat OAuth permissions for any app: review what you grant, prefer minimum-necessary access, and remove plugins you do not use.

What is the difference between a plugin and an agent?

A plugin is a package of capability. An agent is an AI loop that uses capabilities (potentially including plugin-provided ones) to pursue a goal. Plugins are passive (they wait to be called); agents are active (they decide what to do).

Get smarter about AI every morning

Related glossary entries

• Claude Skills
• What is MCP (Model Context Protocol)
• What are AI Agents
• What is Tool Use in AI
• What is Function Calling in AI
• AI Business Glossary
• The Ultimate AI Glossary

Where you might encounter plugins

• Claude for Small Business — includes plugin coverage
• Claude Code Advanced — deep on Claude Code plugins
• Claude Code Beginners Guide
• Microsoft Copilot Guide