AI at Work

Top 10 GitHub repositories for getting more from Claude

The most useful repositories are Anthropic's own: skills, knowledge-work plugins, the official plugin directory, the prompt tutorial, the cookbooks, Claude Code, the Agent SDK and the quickstarts, plus the Model Context Protocol reference servers and one community list of MCP servers. Business users get most from the first four; the rest suit developers.

How we chose these

GitHub is where Anthropic publishes much of its practical material: ready-made skills, plugins, tutorials and code. We checked every repository below on GitHub on 16 November 2026. Eight are Anthropic's, one belongs to the Model Context Protocol project and one is a community list, marked as such.

Star counts are a rough guide to how widely a repository is used, not a measure of quality. The figures below were taken from the GitHub API on 16 November 2026 and rounded.

1. anthropics/skills

Anthropic's public collection of Agent Skills, with about 196,000 stars. It holds the Word, Excel, PowerPoint and PDF skills that power Claude's document features, examples such as brand-guidelines, internal-comms and skill-creator, the Agent Skills specification and a template for writing your own.

  • Who it suits: business users looking for patterns to copy, and developers.
  • How to start: read a few SKILL.md files in the skills folder. Anthropic says the example skills are already available on paid Claude plans. In Claude Code, add it with /plugin marketplace add anthropics/skills.
  • Worth knowing: Anthropic says the skills are for demonstration and education, and should be tested before you rely on them.

2. anthropics/knowledge-work-plugins

Eleven plugins Anthropic built for Claude Cowork, also compatible with Claude Code, with about 36,000 stars. They cover productivity, sales, customer support, product management, marketing, legal, finance, data, enterprise search and bio-research, plus one for building your own plugins. Each is made of markdown and JSON files: skills, slash commands and connector settings.

  • Who it suits: business users on Cowork, and whoever adapts plugins for a team.
  • How to start: install plugins in Cowork from claude.com/plugins. Anthropic describes them as generic starting points that become more useful once you add your own company's context.

3. anthropics/claude-plugins-official

Anthropic's own plugins and the claude-plugins-official marketplace for Claude Code, with about 47,000 stars. Third-party plugins are now submitted through Claude's directory rather than this repository.

  • Who it suits: Claude Code users, technical or not.
  • How to start: in Claude Code, run /plugin and choose Discover, or install one by name with /plugin install followed by the plugin name and @claude-plugins-official.
  • Worth knowing: Anthropic says it does not control what MCP servers, files or other software are included in plugins, and asks you to trust a plugin before installing it.

4. anthropics/prompt-eng-interactive-tutorial

Anthropic's interactive prompt engineering tutorial, with about 39,000 stars. It has nine chapters with exercises, from basic prompt structure to advanced techniques, and an appendix on chaining prompts, tool use and retrieval.

  • Who it suits: anyone who writes prompts for work.
  • How to start: Anthropic recommends the Google Sheets version, which it calls more user friendly, over the code version.
  • Worth knowing: it was written for an older model, Claude 3 Haiku.

5. anthropics/claude-cookbooks

Notebooks and recipes for building with Claude, with about 55,000 stars. It was previously called anthropic-cookbook. Topics include classification, retrieval, summarisation, tool use, images, PDF summarisation, prompt caching, skills and the Agent SDK.

  • Who it suits: developers.
  • How to start: you need a Claude API key. The examples are mainly in Python, and Anthropic suggests its API fundamentals course first.

6. anthropics/claude-code

The public repository for Claude Code, Anthropic's coding tool for the terminal, with about 156,000 stars. It holds the changelog, the issue tracker and example plugins. The full documentation is at code.claude.com.

  • Who it suits: developers, and technical staff who automate work with files.
  • How to start: the README gives a one-line installer for Windows, macOS and Linux, plus Homebrew and WinGet. Installation through npm is deprecated.
  • Worth knowing: the README explains what usage and feedback data Anthropic collects and links to its data usage policy.

7. anthropics/claude-agent-sdk-python

The Python version of the Claude Agent SDK, for building agents on the same tools as Claude Code, with about 8,600 stars. A TypeScript version exists as well.

  • Who it suits: developers building internal tools.
  • How to start: pip install claude-agent-sdk, on Python 3.10 or later. The Claude Code command-line tool is bundled with the package.
  • Worth knowing: by default an agent gets the full Claude Code toolset, including reading, writing and running commands, so decide what it may touch before you run it.

8. anthropics/claude-quickstarts

Starter projects for applications built on the Claude API, with about 21,000 stars. Examples include a customer support agent with a knowledge base, a financial data analyst with charts, and a computer use demonstration.

  • Who it suits: developers, and teams wanting a working prototype to show.
  • How to start: clone the repository, pick a quickstart, set your Claude API key and follow that folder's README.

9. modelcontextprotocol/servers

The reference servers for the Model Context Protocol, the open standard Claude uses to connect to other systems, with about 92,000 stars. The current set is Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time. Older servers such as GitHub, Google Drive and Slack have been archived.

  • Who it suits: developers and IT staff learning how MCP works.
  • How to start: read the Filesystem server, which also underlies Anthropic's Filesystem desktop extension. For servers to use day to day, the project points to the MCP Registry.
  • Worth knowing: the project says these are reference implementations for education, not production-ready solutions.

10. punkpeye/awesome-mcp-servers (community)

A community-maintained list of MCP servers, with about 98,000 stars, grouped by category from databases to workplace tools. A legend marks official implementations and whether a server runs locally or in the cloud. It covers servers you install and run yourself, and carries advertising.

  • Who it suits: developers and IT staff looking for a server for a particular system.
  • How to start: search the list for the system you use, then prefer entries marked as official implementations.
  • Worth knowing: it is not maintained by Anthropic or the MCP project, and inclusion is not a security review.

Before you install anything

  • Read it first. A skill or plugin can contain scripts and instructions to contact outside services. Anthropic advises reading the files of any skill you did not write, including one from a colleague.
  • Check who maintains it. Prefer Anthropic, the MCP project or the vendor of the system you are connecting to.
  • Check what it can touch. Note which folders, accounts and commands each tool can reach, and start with the narrowest access that works.
  • Keep secrets out. Never paste API keys or passwords into a skill, a prompt or a file you commit. Store keys in environment variables or a password manager.
  • Use a test account or test data until a tool has proved itself.
  • On Team and Enterprise, agree with the Claude owner which skills, plugins and connectors are allowed before people install their own.

Choosing and checking tools like these is part of our Claude training in Ireland.

Common questions

Which of these repositories suit someone who does not write code?

anthropics/skills, anthropics/knowledge-work-plugins and the prompt engineering tutorial. The skills and plugins are written in plain markdown, and the tutorial has a Google Sheets version.

What happened to Anthropic's courses repository?

anthropics/courses was archived on 15 September 2026 and is now read-only. Its five courses, including API fundamentals and tool use, can still be read there.

Is the anthropic-cookbook repository the same as claude-cookbooks?

Yes. The old name now redirects to anthropics/claude-cookbooks.

Are community MCP servers safe?

Some are and some are not. A community list does not review the code. Prefer servers built by the vendor of the system you connect to, read the code or have someone technical read it, and give each server the least access it needs.

If you want skills and plugins set up for your team rather than installed one by one, see Claude implementation for Irish businesses.

Sources and product references

Checked 16 November 2026. Features and access can change, so check what your own licence includes.

  1. GitHub: anthropics/skills
  2. GitHub: anthropics/knowledge-work-plugins
  3. GitHub: anthropics/claude-plugins-official
  4. GitHub: anthropics/prompt-eng-interactive-tutorial
  5. GitHub: anthropics/claude-cookbooks
  6. GitHub: anthropics/claude-code
  7. GitHub: anthropics/claude-agent-sdk-python
  8. GitHub: anthropics/claude-quickstarts
  9. GitHub: modelcontextprotocol/servers
  10. GitHub: punkpeye/awesome-mcp-servers (community)
  11. GitHub: anthropics/courses (archived)
  12. Anthropic: use skills in Claude

Editorial guidance informed by Acuity's work with teams. Examples are fictional and client materials stay private. To record who owns an AI use and when it is reviewed, see AI Register.