47 AI Skills, MCP Servers & GitHub Repos That Changed My Workflow

Part 1: Skills & MCP Servers

Skills teach Claude how to do things. MCP gives it access to the outside world. Together they turn Claude Code from a chatbot into an actual workforce. Here are the ones worth installing.

Document & Office (Official Anthropic) 📰

1. PDF Processing— Read, extract tables, fill forms, merge/split. Highest-utility skill for knowledge workers.
   github.com/anthropics/skills → pdf
2. DOCX— Create & edit Word docs with tracked changes, comments, formatting.
   github.com/anthropics/skills → docx
3. PPTX— Slide decks from natural language. Layouts, charts, speaker notes.
   github.com/anthropics/skills → pptx
4. XLSX— Formulas, analysis, charts via plain English.
   github.com/anthropics/skills → xlsx
5. Doc Co-Authoring— Real collaborative writing. Human writes, Claude responds, back and forth.
   github.com/anthropics/skills → doc-coauthoring

Design & Creative 🎨

6. Frontend Design— Escape “AI slop” aesthetics. Real design systems, bold typography. 277k+ installs.
   github.com/anthropics/skills → frontend-design
7. Canvas Design— Social graphics, posters, covers — text in, PNG/PDF out.
   github.com/anthropics/skills → canvas-design
8. Algorithmic Art— Fractal patterns, geometric compositions via p5.js.
   github.com/anthropics/skills → algorithmic-art
9. Theme Factory— Batch-generate color schemes from one prompt.
   github.com/anthropics/skills → theme-factory
10. Web Artifacts Builder— Calculators, dashboards via natural language. No frontend skills needed.
    github.com/anthropics/skills → web-artifacts-builder

Dev & Engineering 👨‍💻

11. Superpowers— 20+ battle-tested skills. TDD, debugging, plan-to-execute pipeline. 96k+ stars.
    github.com/obra/superpowers
12. Systematic Debugging— Root cause analysis first, fix second. 4-phase methodology. S9.2 on SkillHub.
    github.com/obra/superpowers
13. File Search— Ripgrep + ast-grep mastery. S9.0 rated.
    github.com/massgen/massgen
14. Context Optimization— Reduce token costs, KV-cache tricks. 13.9k stars.
    github.com/muratcankoylan/agent-skills-for-context-engineering
15. Skill Creator (Official)— Meta-skill: describe a workflow, get a SKILL.md in 5 min.
    github.com/anthropics/skills → skill-creator
16. Remotion Best Practices— AI video generation. 117k weekly installs.
    github.com/remotion-dev/remotion

Marketing & SEO 🗞

17. Marketing Skills by Corey Haines— 20+ skills: CRO, copywriting, SEO, email sequences, growth.
    github.com/coreyhaines31/marketingskills
18. Claude SEO— Full-site audits, schema validation. 12 sub-skills.
    github.com/AgriciDaniel/claude-seo
19. Brand Guidelines— Encode your brand into a skill. Auto-applies everywhere.
    github.com/anthropics/skills → brand-guidelines

Knowledge & Learning 🧠

20. NotebookLM Integration— Claude + NotebookLM bridge. Summaries, mind maps, flashcards.
    github.com/PleasePrompto/notebooklm-skill
21. Obsidian Skills— By Obsidian’s CEO. Auto-tagging, auto-linking, vault-native.
    github.com/kepano/obsidian-skills
22. Excel MCP Server— Manipulate Excel without Microsoft Excel.
    github.com/haris-musa/excel-mcp-server

All official skills: github.com/anthropics/skills
Browse 80k+ community skills: skillsmp.com

My Must-Have MCP Servers

Beyond skills — these are the ones I love using. Skills teach Claude HOW to do things. MCP gives it ACCESS to the outside world. These three changed my workflow.

Tavily

Search engine built for AI agents. Not blue links — clean structured data. Four tools: search, extract, crawl, map. Connects as remote MCP in one minute.

github.com/tavily-ai/tavily-mcp

Context7

Injects up-to-date library docs into your LLM’s context. No more hallucinated APIs. No more deprecated methods. Just add “use context7” to your prompt. Supports Next.js, React, Supabase, MongoDB, thousands more.

github.com/upstash/context7

Task Master AI

Your AI’s project manager. Feed a PRD → structured tasks with dependencies → Claude executes one by one. Turns chaotic sessions into a proper pipeline. Works across Cursor, Claude Code, Windsurf.

github.com/eyaltoledano/claude-task-master

Part 2: GitHub Repos

25 open-source engines powering the AI revolution. Starting with the most famous and going into more lowkey ones.

Agent Frameworks

23. OpenClaw— The viral AI agent. Persistent, multi-channel, writes its own skills. 210k+ stars.
    github.com/openclaw/openclaw
24. AutoGPT— Full agent platform for long-running tasks.
    github.com/Significant-Gravitas/AutoGPT
25. LangGraph— Agents as graphs. Multi-agent orchestration. 26.8k stars.
    github.com/langchain-ai/langgraph
26. OWL— Multi-agent cooperation. Tops GAIA benchmark.
    github.com/camel-ai/owl
27. Dify— Open-source LLM app builder. Workflows, RAG, agents all-in-one.
    github.com/langgenius/dify
28. CrewAI— Multi-agent with roles, goals, backstories.
    github.com/crewAIInc/crewAI
29. CopilotKit— Embed AI copilots into React apps.
    github.com/CopilotKit/CopilotKit

Local AI

30. Ollama— Run LLMs locally with one command.
    github.com/ollama/ollama
31. Open WebUI— Self-hosted ChatGPT-like interface.
    github.com/open-webui/open-webui
32. LlamaFile— LLM as single executable. Zero dependencies.
    github.com/Mozilla-Ocho/llamafile
33. Unsloth— Fine-tune 2x faster, 70% less memory.
    github.com/unslothai/unsloth

Workflow & Automation

34. n8n— Open-source automation, 400+ integrations + AI nodes.
    github.com/n8n-io/n8n
35. Langflow— Visual drag-and-drop for agent pipelines. 140k stars.
    github.com/langflow-ai/langflow
36. Huginn— Self-hosted web agents. Monitoring, alerts. Privacy-first.
    github.com/huginn/huginn

Search & Data

37. GPT Researcher— Autonomous research → compiled reports.
    github.com/assafelovic/gpt-researcher
38. Firecrawl— Any website → LLM-ready data.
    github.com/mendableai/firecrawl
39. Vanna AI— Natural language → SQL.
    github.com/vanna-ai/vanna

Dev Tools

40. Codebase Memory MCP— Codebase → persistent knowledge graph.
    github.com/DeusData/codebase-memory-mcp
41. DSPy— Program (not prompt) foundation models.
    github.com/stanfordnlp/dspy
42. Spec Kit (GitHub)— Spec-driven dev. Write specs, AI generates code. 50k+ stars.
    github.com/github/spec-kit
43. NVIDIA NemoClaw— Secure sandbox for autonomous agents.
    github.com/NVIDIA/NemoClaw

Curated Collections

44. Awesome Claude Skills— Best curated skill list. 22k+ stars.
    github.com/travisvn/awesome-claude-skills
45. Anthropic Skills Repo— Official reference implementations.
    github.com/anthropics/skills
46. Awesome Agents— 100+ open-source agent tools.
    github.com/kyrolabs/awesome-agents
47. MAGI//ARCHIVE— Daily feed of fresh AI repos.
    tom-doerr.github.io/repo_posts

40 Fresh GitHub Repos Worth Watching

I scanned 1,000+ repos from GitHub forums and picked the most interesting AI projects. Do not forget to do your own security check.

Agent Orchestration & Multi-Agent 🤖

• gstack — Claude Code as virtual engineering team
• cmux — Multiple Claude agents in parallel
• figaro — Orchestrate Claude agent fleets on desktop
• claude-squad — Terminal agents in parallel sessions
• deer-flow (ByteDance) — Sub-agents and sandboxes through skills
• SWE-AF — One API call spins up engineering team
• AIlice — Complex tasks → dynamic agents
• Agent Alchemy — Claude Code + plugins + task manager

Agent Infrastructure & Security 🔐

• Ghost OS — AI agents operate every Mac app
• e2b/desktop — Isolated virtual desktops for agents
• container-use (Dagger) — Containerized environments for coding agents
• Canopy — Encrypted P2P mesh for AI agents
• agent-governance-toolkit (Microsoft) — Security middleware for agents
• claude-code-security-review (Anthropic) — PRs analyzed for security
• promptfoo — Automated security testing for AI models

Memory & Context 📝

• Mem9 — Memory system for AI agents
• Codefire — Persistent memory for coding agents
• Memobase — User profile memory for LLMs

Coding Agents & Dev Tools 💻

• Qwen Code — Terminal AI coding agent by QwenLM
• gptme — Personal AI agent in terminal
• Claude Inspector — See hidden Claude Code prompt mechanics
• TDD Guard — Enforces test-first for AI agents
• rendergit (Karpathy) — Git repo → single file for humans and LLMs
• autoresearch (Karpathy) — Autonomous LLM training system
• pydantic-ai — Type-safe agent framework
• claude-deep-research-skill — 8-phase research with auto-continuation

MCP & Integrations 🔗

• MCP Playwright — Browser automation for LLMs
• stealth-browser-mcp — Undetectable browser automation
• fastmcp — Build MCP servers in minimal Python
• markdownify-mcp — PDFs, images, audio → Markdown
• MCPHub — Manage multiple MCP servers via HTTP

Search, Data & LLM Tools 🔍

• CK (BeaconBay) — Search code by meaning, not keywords
• ExtractThinker — ORM for document intelligence
• OmniRoute — API proxy for 44+ AI providers
• dlt — LLM-native data pipelines from 5,000+ sources
• simonw/llm — Lightweight CLI for local and remote LLMs
• Portkey-AI/gateway — Route requests to 250+ LLMs
• lmnr — Trace and evaluate agent behavior

Video & More 📽

• LTX-Desktop (Lightricks) — Generate and edit videos locally
• MetaClaw — Evolve AI agents without GPU
• Vane — AI answering engine with local LLMs

Where to Find More Skills

• skillsmp.com — 80k+ skills, largest catalog
• aitmpl.com/skills — Templates, one-command install
• skillhub.club — 31k+ skills, AI-rated
• agentskills.io — Official spec

Install any skill:
Personal: ~/.claude/skills/
Project: .claude/skills/
Clone, copy, restart. Done.