Claude Plugins in 2026: The Complete Practitioner Guide
If you have heard of Claude plugins but are not sure what they actually do, how they differ from MCP servers, or where to even find them, you are in the right place. According to Claude.ai documentation, plugins are capability packages that can combine MCP connectors, skills, slash commands, and sub-agents into a single installable unit, which means one plugin can do far more than one integration ever could on its own. I have spent the better part of 2026 testing these across real workflows, and this guide covers everything I wish someone had told me before I started.
Key takeaways
- What are Claude plugins? Installable capability packages for Claude that bundle MCP connectors, skills, slash commands, and sub-agents together, extending what Claude can do inside the Claude desktop app.
- Plugins vs. MCP servers. MCP servers are the protocol layer that brings context into Claude. Plugins are the distribution and packaging unit that wraps one or more MCP connections into something you can install in two steps.
- Where to install. The official Claude plugins marketplace is surfaced directly inside Claude Code and claude.ai, with an official directory automatically available to all users, per Anthropic’s plugin submission documentation.
- Claude Code plugins are a specific category designed to extend the coding assistant experience, including custom Claude Code commands, hooks, and project-specific tooling.
- The CLAUDE.md file matters. Many plugins read from your CLAUDE.md file configuration to personalize behavior per project, so setup is a one-time investment rather than a repeated configuration task.
- Availability depends on your MCP type. If a plugin references a remote MCP, it works on web and desktop. If it references a local MCP, it is limited to Claude Desktop and Claude Code.
- Build your own. Anthropic documents a packaging path where you can bundle an MCP-based local server into a plugin using
.mcp.jsonand submit it to the plugin directory, per Claude’s connectors documentation.
What are Claude plugins and why should you care?
The simplest way I can describe Claude plugins is this: they are the moment Claude stops being a smart chat interface and starts being an actual work tool.
Without plugins, Claude responds to prompts. With plugins, Claude can pull live data, run tasks in connected tools, execute code, write files to your system, and hand off work to Claude Code subagents, all within the same conversation thread.
Plugins package multiple capabilities together. A single Claude plugin might include an MCP connector to your project management tool, a set of slash commands to trigger specific workflows, and a sub-agent configuration that handles a recurring approval step. You install it once. It runs every time you need it.
This is very different from what most people mean when they compare Claude to ChatGPT. ChatGPT plugins were primarily about fetching web results or calling one external API. Claude plugins in the desktop app are closer to installable workflow automations that live inside your AI assistant.
For anyone building a serious workflow on top of Claude, understanding plugins is not optional. It is the foundation.
Claude plugins vs. MCP servers vs. skills: the real difference
This is the question I get asked more than any other, and the answer is cleaner than most people expect.
MCP servers (the Model Context Protocol layer) are the protocol surface. They are background servers that make tools, resources, and context available to Claude. You can think of MCP servers as the plumbing. They define what data and actions Claude can access, but they do not handle how that capability is packaged, distributed, or discovered by end users.
Skills are more granular. A skill is typically a single callable capability, like the ability to search a specific database or run a particular transformation. Skills live inside plugins but are not the same as plugins.
Claude plugins are the container. One plugin can reference one or several MCP servers, wrap a set of skills, include Claude Code commands, and configure sub-agent behavior. When you install a plugin from the Claude plugins marketplace, you are installing all of that at once.
A useful mental model: MCP is the API layer. Skills are the functions. Plugins are the application. You can use MCP servers directly if you are technical and want granular control, but plugins are what most operators and founders should reach for first.
The decision point documented at Anthropic’s “What to build” guidance is clear: if a plugin references a remote MCP URL, it works across web, desktop, and Claude Code. If it references a local MCP, it is available in Desktop and Claude Code only. That single decision shapes your entire team rollout strategy.
The official Claude plugins marketplace and install flow
The official Claude plugins marketplace is not a separate website you visit. It is surfaced inside Claude Code and in your claude.ai settings panel, and it is automatically available to all users, per Anthropic’s plugin submission documentation.
Here is the install flow I use:
- Open Claude Code or navigate to your claude.ai settings.
- Go to the Plugins section in the sidebar.
- Search the directory by category or keyword.
- Click Install on the plugin you want. For command-line installs inside Claude Code, the format is
/plugin install {plugin-name}@claude-plugins-official, documented at the official claude-plugins-official repository on GitHub. - Authorize any permissions the plugin needs (file access, external API connections, etc.).
- The plugin appears in your active sessions immediately.
One thing worth knowing: if a plugin includes an MCP URL that is not in Anthropic’s official directory, it will show up as “Custom” in your settings, per Claude’s connector documentation. That is not a problem for personal use, but it does create extra friction for team onboarding. If you are rolling out plugins across an organization, stick to directory-listed plugins wherever possible.
For team and enterprise contexts, plugin management gets its own governance layer. You can restrict which plugins are available to specific users, set approval workflows before installation, and audit usage through the Analytics API. Anthropic provides an Analytics API that returns token usage and USD cost data across Claude surfaces (currently in beta), per the Claude Enterprise Analytics API documentation. That cost visibility matters a lot once your team is running a dozen active plugins simultaneously.
Claude Code plugins: what changes when you code with plugins
If you spend time in Claude Code, the plugin model changes your experience more than it does for casual claude.ai users.
Claude Code plugins can add capabilities that pure chat interactions cannot touch. That includes:
- Custom Claude Code commands that trigger multi-step workflows (lint, test, deploy) with a single slash command.
- Claude Code hooks that fire automatically when specific events occur, like a file save or a failed test.
- Claude Code subagents that run parallel workstreams inside a plugin’s execution environment, processing tasks concurrently rather than sequentially.
- Claude Code plan mode integrations that let a plugin see the full execution plan before any code runs, allowing for pre-flight checks or approval gates.
This is one of the main reasons I find the Claude Code vs. Cursor comparison so interesting in 2026. Cursor has a strong code completion model, but Claude Code with plugins installed is running a fundamentally different architecture, one that can coordinate multiple agents, write to your file system, and execute tasks in an isolated environment, per Claude’s plugins overview.
Plugin categories with concrete examples
The Claude plugins marketplace organizes plugins into categories. Here is how I think about each one and what I have seen work in practice.
Productivity and project management
These plugins connect Claude to tools like Linear, Notion, or Jira. Instead of copying project context into a prompt manually, the plugin surfaces the right tickets, meeting notes, or specs directly inside your Claude session. Claude can then draft updates, flag blockers, and generate status summaries without you switching tabs.
Code and development
Claude Code plugins in this category add things like repository awareness, automated test runners, and deployment triggers. The best ones use Claude Code hooks to fire without manual prompting, so your workflow stays uninterrupted. If you have set up Claude Code commands in your project, a good development plugin will pick up on those command patterns and extend them.
Legal and compliance
Anthropic has been actively expanding this category in 2026. These plugins add document review capabilities, contract clause extraction, and compliance checklist generation. They use sub-agents to process long documents in parallel, which cuts review time significantly. The execution happens inside an isolated environment, so sensitive documents are not written to shared storage unless you configure them to be.
Data and analytics
These connect Claude to databases, BI tools, and spreadsheet sources. The MCP layer handles the query execution, while the plugin presents results in a format Claude can reason over. This is where I have seen the biggest time savings, because the alternative is manually pulling exports, formatting them, and then pasting context into a prompt.
Memory and context
Some plugins work directly with your CLAUDE.md configuration to maintain context across sessions. They read from your CLAUDE.md file, store project-specific preferences, and surface relevant history without requiring you to re-explain your setup every time you open a new session. Proper CLAUDE.md configuration is what makes these plugins work well, so if you have not set that up yet, that guide is worth reading first.
Communication and writing
These plugins connect Claude to email clients, CRMs, and publishing tools. For founder-level work, I find these most useful for first drafts and inbox triage, not final outputs. The plugin handles the mechanical parts, and I handle judgment calls.
How to choose Claude plugins for your work
The honest answer is that most people install too many plugins too fast. Here is the framework I actually use.
Step 1: Identify your highest-repetition tasks. Plugins are packaging for frequently-run tasks, as Anthropic’s documentation describes them. If a task only comes up once a month, a plugin adds more overhead than value. If you do something ten times a week, that is your first plugin candidate.
Step 2: Check whether you need remote or local MCP. If you want a plugin available across your web sessions AND Claude Code, choose one that references a remote MCP server. If you only use Claude Code on one machine, a local MCP plugin gives you more control. The Claude Help Center article on desktop vs. web connectors lays this out clearly.
Step 3: Check the directory status. If a plugin references an MCP URL not in Anthropic’s official directory, it shows as “Custom” in your settings. That matters for team rollouts because onboarding friction goes up. For personal use, it is a non-issue.
Step 4: Set the right cross-tool expectations. People who switch from ChatGPT sometimes expect plugin behavior to mirror what they knew from that ecosystem. Claude plugins operate differently, particularly around sub-agent coordination and file system access. Setting the right expectations before installing saves troubleshooting time later.
Step 5: Watch your costs. Once you have enterprise access to the Analytics API, monitor token usage per plugin. Some plugins fire MCP calls on every message, which compounds cost quickly if you have long sessions. The Analytics API returns USD cost data per surface, so you can compare plugin value against actual spend.
Build your own Claude plugin: a preview
If you have an internal tool that your team uses constantly, building a custom Claude plugin is more approachable than it sounds.
The basic path documented at Claude’s connectors overview works like this:
- Start with a local MCP server that exposes your internal tool’s API as a set of callable tools.
- Test it in Claude Code or Claude Desktop first, using direct MCP configuration.
- Once it works reliably, bundle it into a plugin using
.mcp.jsonas the packaging manifest. - Define the plugin’s metadata (name, description, permissions, category).
- Submit it to the official Claude plugins directory if you want automatic availability, or distribute it as a custom plugin internally.
The distinction between remote and local MCP becomes critical at step 3. If your internal tool runs on a server your team accesses remotely, package it with a remote MCP URL so it works across all Claude surfaces. If it runs on individual machines (a local database, a proprietary CLI tool), package it as a local MCP plugin and expect Claude Code and Desktop deployment only.
Claude Code plan mode is particularly useful during plugin development because it lets you see the full execution plan before committing. That makes it easier to catch unintended side effects before they affect real data.
For organizations with specific compliance requirements, the isolated execution environment matters here. Plugins run tasks in parallel workstreams and write outputs to your file system, but they do so within defined boundaries. That architecture is what makes Claude plugins safe to deploy in regulated industries.
I also recommend configuring your CLAUDE.md file before finalizing any custom plugin, because many plugin behaviors can be tuned at the project level through that file. The memory architecture and the plugin architecture interact, and getting both right is what separates a frustrating setup from a genuinely useful one.
Claude plugins in team and enterprise contexts
Installing plugins for yourself is one thing. Rolling them out to a team introduces a different set of considerations.
The first one is governance. When a team member installs a Claude plugin that includes external API access, that access runs under their credentials unless you configure it otherwise. In an enterprise setting, you want organization-level plugin management turned on, so admins control the approved plugin list rather than leaving it to individual users.
The second is cost attribution. This is where the Analytics API becomes genuinely valuable, not just interesting. When I can see token usage and USD cost broken down by surface and session, I can make real decisions about which plugins earn their spot in the workflow and which ones are adding cost without adding value.
The third is the cross-tool expectation conversation that inevitably happens during rollout. Teams with ChatGPT experience often underestimate how much more coordination work Claude plugins support. Claude Code subagents running in parallel, hooks firing on triggers, and plan mode providing pre-run visibility are all capabilities that require some onboarding investment to use well. Build that training time into your rollout plan.
For organizations evaluating Claude Code against other tools, the Claude Code vs. Cursor comparison breaks down the specific capability differences that affect team adoption decisions.
Frequently asked questions
What are Claude plugins and how do they work?
Claude plugins are installable capability packages that extend what Claude can do inside the Claude desktop app. A single plugin can bundle MCP connectors, skills, slash commands, and sub-agents together, so installing one plugin can activate multiple connected capabilities at once, per Claude.ai documentation.
What is the difference between Claude plugins and MCP servers?
MCP servers are the protocol layer that brings tools and context into Claude. Claude plugins are the packaging and distribution layer that wraps one or more MCP connections into something you can install from the official directory. You can use MCP servers directly, but plugins make that process accessible without manual configuration.
Where is the Claude plugins marketplace?
The official Claude plugins marketplace is surfaced inside Claude Code and in your claude.ai settings panel. It is not a standalone website. Anthropic describes the directory as automatically available to all users, with official plugins appearing by default and custom plugins appearing when the referenced MCP URL is not in the directory.
How do I install Claude Code plugins?
Inside Claude Code, use the command /plugin install {plugin-name}@claude-plugins-official for official plugins, as documented at the official claude-plugins-official repository on GitHub. You can also install through the Claude Code settings UI by browsing the plugin directory and clicking Install.
Do Claude plugins work the same on web and desktop?
It depends on the MCP type inside the plugin. If the plugin references a remote MCP server, it works across web, Claude Desktop, and Claude Code. If it references a local MCP, it is limited to Claude Desktop and Claude Code only, per the Claude Help Center. Choose remote MCP plugins when cross-surface consistency matters.
Is it worth building a custom Claude plugin in 2026?
Yes, if your team runs a specific internal tool frequently. Anthropic documents a clear packaging path using .mcp.json that lets you bundle a local MCP server into a submittable plugin. The build investment pays off fastest for high-repetition workflows where manual context switching is currently eating time every day.
How do Claude plugins interact with the CLAUDE.md file?
Many plugins read from the CLAUDE.md file at the project level to personalize behavior, pick up project-specific conventions, and avoid re-prompting you for context. Getting your CLAUDE.md file configuration right before installing plugins means those plugins work correctly from the first session rather than requiring repeated correction.
The shift from reading about AI to actually using it
I have watched a lot of people spend months reading about Claude, setting up accounts, exploring settings, and never quite crossing into consistent daily use. In almost every case, the missing piece was not knowledge. It was a specific workflow that made Claude indispensable for something they already needed to do.
Claude plugins are that bridge. When you install a plugin that connects directly to your project management tool, your codebase, or your document library, Claude stops being a general AI you interact with occasionally and starts being a specific tool that sits inside your actual work. The Claude Code plugins for development, the memory plugins that read your CLAUDE.md file, the connectors that surface live data through MCP servers, all of that infrastructure exists to make Claude functional for real tasks, not just impressive in demonstrations.
This is the moment the technology shifts from something you read about to something you actually use to do work.
If you are ready to build a more structured workflow around Claude, the next step is to give Claude the project context, voice, and workflows that make it yours. The Claude AI Strategy Stack at the Claude plugins at Strategized.ai walks you through a 30-minute guided intake that does exactly that. One configured workspace, one Claude that knows your work.
Plugins only work when Claude itself is online, so if yours suddenly stop responding, rule out an outage first: check whether Claude is down.
Skip the build, install the plugin
This guide covers how plugins work. If you would rather not build one, there are 12 ready for specific roles — marketing agencies, consultants, accountants, real estate, and more — each with 20+ capabilities already written.
Browse the 12 plugins →12 roles available. $99 one-time each at Strategized.ai/plugins, or All-Access at $797, and they run on the free Foundation Plugin.



