If Cowork is the workspace where complex tasks get done, Plugins are the specialists that show you how each task should be done.
Think of it like a well-equipped kitchen. Cowork gives you the counter space, the knives, the heat source — the environment where real work happens. Plugins are the recipe books written by people who’ve cooked in that kitchen before. Each one carries domain knowledge, a preferred sequence, and a shortcut for the steps you’d otherwise figure out by trial and error. You can follow them as written, adapt them, or write your own from scratch.
Anthropic launched Plugin support for Cowork on January 30, 2026, and shipped a major enterprise update on February 24. In six weeks, they’ve gone from 11 open-source plugins to a full marketplace with private org-level distribution and cross-app workflows between Excel and PowerPoint. The pace is aggressive, and it tells you where Anthropic is betting.
This article is about what those plugins can actually do — and the specific gap I found when I tried to use them for a real decision.
What we will cover today:
Data Analysis Plugin — auditing my publication with real data
Marketing Plugin — campaign plan, competitive brief, and content drafting
What I found, what’s next, and early access to the beta tool 🚀
What makes up a Plugin
A plugin bundles four components into one installable package: Skills (domain knowledge Claude draws on automatically in the background), Slash commands (explicit actions you trigger — type / and pick a workflow like /campaign-plan or /competitive-brief), Connectors (MCP links to external tools like HubSpot, Ahrefs, or your CRM), and Sub-agents (parallel workers that split complex tasks so you’re not waiting for sequential execution).
Everything is file-based — markdown and JSON, no code — which means anyone can read, edit, and share plugins.
Quick context: where this picks up from
Two weeks ago, I published a piece on using Cowork to build a complete newsletter content pipeline — four input files in, three publish-ready outputs out, one session. That article covered the base Cowork environment: direct file access, autonomous multi-step execution, mid-task steering, and how to structure your working folder.
New to Claude and want to catch up with all it’s features? I’d suggest reading Jenny Ouyang’s piece on How to Onboard to Claude Without the Learning Curve, or Daria Cupareanu’s on How I Switched from ChatGPT to Claude Without Losing Data if you are migrating to Claude from ChatGPT.
If Cowork provides the rails, Plugins provide the engine. This piece is about what happens when you install that engine and push it on a real decision — not a demo, not a test, but something I actually needed to figure out.
Here’s how you can access any Anthropic Plugin in Cowork:
One distinction worth making early: a Skill is a single instruction set that shapes how Claude approaches one domain. A Plugin bundles multiple skills together with slash commands, connectors, and sub-agents into a complete workflow package — the difference between one ingredient and the full recipe. Many of the specificities in Claude are very well explained in Karo (Product with Attitude)’s Claude Cowork Guide for Power Users: 50+ Tested Tips on Plugins, Skills, Sub-Agents, and Memory.
The Plugin ecosystem: what’s available today 🔌
Anthropic shipped 11 open-source plugins at launch, covering sales, marketing, finance, legal, data analysis, engineering, product management, customer support, enterprise search, productivity, and a meta-plugin for building your own. Here are the some relevant to my workflow, with my one-line take on each.
The full list of all 11 is on Anthropic’s GitHub. All are open-source and customizable.
Two observations from browsing the lineup.
These are starting points, not finished products — the real power comes when you customise them for your specific tools, terminology, and processes.
Slash commands are the real differentiator from standalone Skills. A Skill shapes how Claude thinks. A slash command triggers a complete workflow in one invocation. That’s a meaningfully different experience.
If you’ve used base Cowork (file access, multi-step execution, mid-task steering), plugins add three things on top: pre-loaded domain knowledge so you don’t re-explain context every session, slash commands that trigger complete workflows in one invocation, and connectors that reach external tools beyond your local filesystem.
The trade-off: plugin sessions consume quota faster — a complex multi-command session can burn through the equivalent of 100+ standard messages.
The question I needed to answer
Let’s move into our main discussion in this article.
Every week, we hear about a new AI tool or a major upgrade to an existing one. Cursor 2.0, Gemini 3.1 Pro, Seedance 2.0, GPT-5.3-Codex — the list from February alone would take an afternoon to research properly.
For Cash & Cache, this isn’t just curiosity. Every tool we evaluate is a potential article, a potential workflow change, or a potential recommendation to our readers. Getting it wrong — hyping something that doesn’t deliver, or missing something that does — costs credibility.
So I had a specific question: Is Cursor worth adopting for my workflow? Not “is it a good product” in the abstract, but specifically — given how I work, what I produce, and who I produce it for — does switching from VS Code + GitHub Copilot to Cursor make practical sense?
I wanted to see whether Cowork Plugins could help me answer that. Not one plugin in isolation, but a sequence: use the Data Analysis plugin to understand my current workflow, then use the Marketing plugin to plan how I’d communicate the decision, and see how far the pre-built ecosystem gets me before I need something custom.
Here’s where we get into the implementation.