GitWatch

The fastest-growing GitHub repo in history — 9K to 247K stars in weeks (18× faster than Kubernetes). Free, open-source autonomous AI agent that sends emails, writes code, monitors servers, and orchestrates multi-agent workflows.

The reference implementation for autonomous AI agents. Understanding OpenClaw is understanding the architecture pattern every agentic system is converging on.

Official OpenClaw skill directory — 5,700+ community-built skills for every workflow. The npm registry of the agentic era.

The central registry for OpenClaw capabilities. If you want to extend what an agent can do, this is where you start.

Curated list of 3,009 high-quality OpenClaw skills — the quality filter on top of ClawHub's 5,700+ raw registry.

The curated shortlist for serious builders — saves hours of evaluation by surfacing the skills that actually work reliably in production.

Typed workflow runtime for OpenClaw. Define complex multi-step agentic tasks as type-safe, composable workflows.

The missing layer between "run a skill" and "orchestrate a complex multi-agent pipeline" — how production OpenClaw workflows are built.

Security layer for OpenClaw — permission manifests, runtime enforcement, sandboxing, and audit logging.

With 20% of ClawHub plugins flagged as potentially malicious, ClawGuard is the enterprise prerequisite before any production OpenClaw deployment.

Smart LLM router for OpenClaw — routes tasks to the optimal model across 30+ providers. Saves 78% on inference costs.

At scale, model routing is the difference between a viable and unviable cost structure. 78% savings is not marginal.

Minimalist OpenClaw implementation for Telegram. Ultra-lightweight, single-file deployment — the fastest way to get an OpenClaw agent running.

The fastest path from zero to running agent. If you want to understand OpenClaw architecture without the full stack, start here.

OpenClaw principles implemented in C on ESP32-S3 microcontrollers. Runs a full agentic loop on a $5 chip.

Proof that the agentic architecture is not cloud-dependent. Edge AI agents on embedded hardware is the next frontier.

MCP tool access to MarkItDown — converts any file format to Markdown for LLM consumption.

The universal file ingestion layer for any Claude workflow that needs to read documents.

The official MCP reference servers by Anthropic. The canonical starting point for all MCP development.

Every MCP server is built against this spec. Understanding it is understanding the protocol.

Up-to-date code documentation for LLMs. Pulls live, version-specific docs into Claude's context.

Eliminates the #1 cause of AI coding errors — stale, hallucinated documentation.

Connect and unify data across 200+ platforms and databases through a single MCP server.

Single MCP endpoint for all your data sources — no more per-database connectors.

Official Microsoft Playwright MCP server. Enables LLMs to interact with web pages via accessibility snapshots.

The gold standard for giving Claude a browser — accessibility-first, not fragile CSS selectors.

Official GitHub MCP server. Full integration with repos, PRs, issues, code search, and more.

Lets Claude operate your entire GitHub workflow without leaving the chat.

AI-powered task management for AI-driven development. PRD parsing, task expansion, dependency tracking.

Turns a product requirements doc into a structured, executable task tree Claude can work through.

A fully-featured coding agent using language servers (LSP) for symbolic code operations.

The most capable open-source coding agent for Claude — uses real language server intelligence.

MCP server for Blender. Claude creates and manipulates 3D scenes, objects, and materials via natural language.

The breakout creative tool MCP — proof that agentic workflows extend into 3D production.

Local-first screen/audio capture with semantic search and LLM-powered history analysis.

The most ambitious personal AI memory system — gives Claude total recall of everything on your computer.

MCP server for Figma. Claude reads design files, inspects components, and generates code from designs.

Closes the design-to-code gap — Claude can read a Figma file and write the implementation directly.

TypeScript framework for building MCP servers with minimal boilerplate. The fastest way to ship a custom server.

If you want to build your own MCP server, this is the starting point — not the official SDK.

Official AWS MCP servers. Claude can query AWS documentation, manage resources, and interact with AWS services.

The official AWS integration — gives Claude native access to the entire AWS service catalog.

Community Playwright MCP server with extended test automation capabilities and screenshot analysis.

The community alternative to Microsoft's official Playwright MCP — more features, faster iteration.

Cloud browser automation via Browserbase. Run headless browsers at scale without local infrastructure.

Moves browser automation to the cloud — no local Chrome instance, no resource conflicts, infinitely scalable.

Official visual debugging tool for MCP servers. Inspect tools, resources, and prompts in a browser UI.

The essential development companion — you cannot effectively build MCP servers without this.

Production-grade Python agent framework from the Pydantic team. Type-safe, testable, multi-model.

The most production-ready Python agent framework — built by the team behind the most-used Python library.

Google's open-source AI agent for the terminal. Comparable to Claude Code — uses Gemini 2.5 Pro.

The primary competitor to Claude Code — worth understanding the alternative approach.

MCP server for Supabase. Claude can query databases, manage tables, and interact with Supabase projects.

Supabase is the most popular open-source Firebase alternative — this gives Claude native access to it.

High-performance code editor with native MCP support and deep Claude integration built in.

The fastest code editor with the best native AI integration — Claude is a first-class citizen, not a plugin.