Anthropic publishes a substantial amount of open-source code on GitHub. For builders shipping on Claude, knowing this catalog saves real time — the SDK that makes Anthropic API integration trivial, the MCP servers that drop into your Claude Code config, the plugins library, the skills repo, the cookbook of working patterns. This post is the indexed builder's tour as of May 2026.
The org page
github.com/anthropics is the canonical org. Bookmark it. Repositories are added regularly — checking the org page monthly is the easiest way to discover new tooling.
Claude Code & agent tooling
- anthropics/claude-code — the Claude Code CLI. Issues here are the right place to report bugs in the terminal agent. Releases page tracks updates.
- anthropics/claude-code-action — GitHub Action that runs Claude Code against your repo in CI. Use this to have Claude review PRs, run autonomous tasks on a schedule, or respond to issues.
- Agent SDK (in the SDK repos below) — higher-level agent abstractions on top of the raw API.
SDKs
- anthropic-sdk-python — Python SDK. The standard for backend integrations.
- anthropic-sdk-typescript — TypeScript / Node SDK. What your Railway backend likely uses.
- anthropic-sdk-go — Go SDK.
- anthropic-sdk-java / kotlin / ruby — community-supported in some cases, official in others; check current status.
- anthropic-sdk-swift — if you're calling Claude from iOS Swift code (which you should NOT do directly — route through your backend), this exists for server-side Swift uses.
MCP servers and protocol
Model Context Protocol is the open standard Anthropic launched for connecting Claude to external tools. Anthropic maintains a reference suite of MCP servers:
- modelcontextprotocol/servers — reference MCP servers maintained by Anthropic and contributors. Servers include: filesystem, fetch, git, GitHub, GitLab, Google Drive, Postgres, SQLite, Puppeteer (browser automation), Sentry, Slack, time, brave-search, memory, etc.
- modelcontextprotocol/specification — the protocol spec itself.
- modelcontextprotocol/python-sdk — build custom MCP servers in Python.
- modelcontextprotocol/typescript-sdk — build custom MCP servers in TypeScript.
- modelcontextprotocol/registry — the discovery registry for community MCP servers.
For an iOS dev using Claude Code, the most useful MCP installs are: filesystem (built in), GitHub, fetch (web requests), and any vendor-specific ones for your stack (Railway, App Store Connect, etc.).
anthropics/claude-plugins-official
anthropics/claude-plugins-official is Anthropic's officially maintained set of Claude plugins — bundled capabilities that drop into Claude Code or Claude.ai to extend functionality. Categories typically include: developer workflows (code review, refactor agents), content (writing assistants), data (analytics agents), and admin (project management).
Installing a plugin is generally one command in Claude Code; the repo's README will have current instructions.
anthropics/skills
anthropics/skills is Anthropic's reference repository for Claude Skills — reusable instruction packs that Claude can load on demand. A skill is typically a Markdown file with a description, when-to-use guidance, optional tools, and example invocations.
Skills are the 2026 evolution of "system prompt templates" — instead of stuffing your system prompt with every possible instruction, you ship skills that Claude loads contextually when needed. Reduces context bloat, sharper specialization.
Community repositories like agentskills.io, skillsmp.com, and others publish third-party skills. We cover the ecosystem in detail in our Claude Skills Ecosystem post.
Cookbook & examples
- anthropic-cookbook — working code examples and patterns: tool use, RAG, vision, structured outputs, sub-agents, fine-tuning, evaluations. The fastest path to "how do I do X with Claude" for most X.
- prompt-eng-interactive-tutorial — Anthropic's official prompt engineering course as Jupyter notebooks. Genuinely the best place to learn prompting.
- anthropics/courses — broader course materials including agent building, evaluations.
Research, evaluations, and supporting tools
- anthropics/evals — evaluation frameworks and datasets for measuring model performance.
- anthropic-long-context-prompting — patterns specifically for using long-context windows effectively (relevant when you're using Opus 4.7 1M).
- Research repos — periodic releases tied to safety / alignment / interpretability papers. Check the org page for current ones.
Community / how to engage
- Issues on relevant repos are read by Anthropic engineers. Reproducible bug reports get traction.
- Discussions tab on the larger repos is where best practices emerge.
- Forks in your account let you customize tooling without losing upstream updates.
- Pull requests — small improvements (docs fixes, examples, edge-case handling) are welcomed.
For a working iOS dev shipping on Claude, the four repos to bookmark are: anthropic-sdk-typescript (your backend), claude-code (your IDE), anthropic-cookbook (when stuck on "how do I do X"), and modelcontextprotocol/servers (when you want to add a capability to your agent).
See also: Claude at Maximum Efficiency, Agentic AI in Claude, Claude Skills Ecosystem, GitHub for Beginners.
- github.com/anthropics — the org page
- docs.anthropic.com — official documentation
- modelcontextprotocol.io — MCP spec