Abstract: Suitable for developers, content creators, product builders, and beginners.

Editor's Note: If you've been following AI recently, it's easy to get the illusion: there are more and more tools, but fewer and fewer that you can actually use.

For most people, the difficulty is not whether there are tools, but which tools are worth using and where to start.

This list is based on practical testing and systematically sorts the current AI tool ecosystem, filtering out 60 tools and categorizing them by different levels and usage scenarios: from coding and development tools, to agent frameworks and tool integration, to workflow automation, data processing and infrastructure, and finally extending to learning resources and continuous update paths.

Overall, it outlines a clear path from "Using AI" to "Building AI systems," providing different entry paths for developers, content creators, product builders, and beginners, helping readers find a suitable entry point in the complex tool ecosystem.

In an era of constant tool iteration, such a structured overview is perhaps more meaningful than single-point recommendations. It provides not answers, but a more efficient way to understand and use AI.

The original text is below:

I spent over 100 hours testing various AI tools so you don't have to step on the pitfalls yourself.

Bookmark this first :)

By 2026, the AI tool ecosystem has reached a level of "information overload": new frameworks every week, new agents every day, new projects topping GitHub's trending list every morning.

Most of it is just hype. Some are indeed useful. A very few will truly change how you work.

I've filtered out the noise for you.

The 60 tools below are those I've personally tested and filtered out as currently worth paying attention to—organized by category, along with the scenarios they are truly good at.

Suggest you bookmark it; you'll likely come back to use it repeatedly.

Part 1: AI Programming Agent & IDE

These types of tools allow AI to help you write code, review code, and even manage the entire development process.
The key point: these are ones that can actually run in real workflows, not just stay in demos.

1. Claude Code
A command-line programming agent launched by Anthropic. Can read files, write code, run tests, and operate directly in your local environment.
If you want to maintain full control in AI-assisted development, this is currently the "gold standard".
https://docs.anthropic.com/en/docs/claude-code

2. Cursor
An AI-first code editor built on VS Code. Supports inline completion, conversation with the codebase, and multi-file editing.
Suitable for developers who want to seamlessly integrate AI into existing development workflows.
https://www.cursor.com

3. Codex CLI
A terminal programming agent launched by OpenAI. Give instructions in natural language, it will read the codebase, generate and execute code.
Very strong in multi-step task implementation.
https://github.com/openai/codex

4. Windsurf
An AI programming IDE launched by Codeium. Its Cascade agent supports multi-file editing, deep understanding of the codebase, and an "immersive coding flow".
Growing fast.
https://codeium.com/windsurf

5. Superpowers
A collection of 20+ practical skills built for Claude Code, including TDD, debugging, pipelines from planning to execution, etc.
Over 96k stars on GitHub. If you use Claude Code, it is recommended to install this first.
https://github.com/obra/superpowers

6. Spec Kit (GitHub)
A "Specification-Driven Development" tool: write the specification first, then have the AI generate the code. Forces you to think clearly before starting. Over 50k stars on GitHub.
https://github.com/github/spec-kit

7. Aider
An AI pair programming tool in the terminal, supports any LLM.
Especially strong in handling existing codebases. Over 30k stars on GitHub.
https://github.com/paul-gauthier/aider

Part 2: Agent Frameworks

Used to build automated systems that can "think-act-iterate".

8. OpenClaw
A phenomenal open-source AI agent. Supports long-running operation, multi-channel (WhatsApp / Telegram / Discord), and can even write its own skills.
Over 210k stars on GitHub, currently one of the easiest entry points for personal AI agents.
https://github.com/openclaw/openclaw

9. LangGraph
Uses "graph structures" to orchestrate multiple agents: supports branching logic, human-in-the-loop, and persistent state.
https://github.com/langchain-ai/langgraph

10. CrewAI
A multi-agent collaboration framework where each agent has a role, goal, and "persona".
Suitable for simulating team collaboration processes.
https://github.com/crewAIInc/crewAI

11. AutoGPT
A veteran fully automated agent framework, suitable for long-running tasks.
Much more mature compared to early versions.
https://github.com/Significant-Gravitas/AutoGPT

12. Dify
An open-source LLM application building platform that integrates workflow, RAG, agents, and model management.
Also friendly to non-developers.
https://github.com/langgenius/dify

13. OWL
A multi-agent collaboration framework, leading in GAIA benchmark performance.
Represents cutting-edge research moving towards practical use.
https://github.com/camel-ai/owl

14. CopilotKit
Allows embedding AI copilots directly into React applications.
Not just improving development efficiency, but making AI part of the product.
https://github.com/CopilotKit/CopilotKit

15. pydantic-ai
A type-safe agent framework based on Pydantic.
Suitable for Python developers who want structured, verifiable outputs.
https://github.com/pydantic/pydantic-ai

Part 3: MCP Services & Tool Integration

MCP (Model Context Protocol) allows AI to truly 'connect to the world'. Skills teach it *how* to do things, MCP gives it *the permission to do* them.

16. Tavily
A search engine designed specifically for AI agents, not blue links, but structured data that can be used directly by LLMs.
Offers four tools: search, extract, crawl, and maps; can be integrated with MCP in one minute.
https://github.com/tavily-ai/tavily-mcp

17. Context7
Injects the latest library documentation into the LLM's context.
No more "hallucinated APIs" or outdated methods.
Add a line "use context7" in the prompt to automatically pull the latest docs.
https://github.com/upstash/context7

18. Task Master AI
Your AI project manager. Input a PRD, it breaks it down into tasks with dependencies.
Then executed step-by-step by Claude, turning a chaotic development process into an orderly pipeline.
https://github.com/eyaltoledano/claude-task-master

19. MCP Playwright
Provides browser automation capabilities for LLMs.
Can control a real browser with natural language: testing, scraping, interaction can all be done.
https://github.com/executeautomation/mcp-playwright

20. fastmcp
Quickly build MCP services with minimal Python code.
One of the fastest paths to create custom tool integrations for Claude and other models.
https://github.com/jlowin/fastmcp

21. markdownify-mcp
Converts various formats like PDF, images, audio, etc., into Markdown.
Allows any document to enter the AI workflow.
https://github.com/zcaceres/markdownify-mcp

22. MCPHub
Manages multiple MCP services via HTTP.
A single panel to uniformly manage all tool connections.
https://github.com/samanhappy/mcphub

Part 4: Claude Skills (Curated)

Skills can inject "professional workflow capabilities" into Claude. The community currently has 80k+ skills; the ones below are truly worth installing.

23. PDF Processing (Official)
Supports reading PDFs, extracting tables, filling forms, merging and splitting files.
For knowledge workers, one of the most practical skills.
https://github.com/anthropics/skills/tree/main/skills/pdf

24. Frontend Design (Official)
Used to build truly usable design systems, including bold typography and production-ready UI.
Can break away from the common "AI-generated" design style. Over 277k+ installations.
https://github.com/anthropics/skills/tree/main/skills/frontend-design

25. Skill Creator (Official)
A "meta-skill". Describe your workflow in natural language, generate a complete SKILL.md within 5 minutes.
No need to write configuration to create new skills.
https://github.com/anthropics/skills/tree/main/skills/skill-creator

26. Marketing Skills (Corey Haines)
Covers 20+ skills including CRO, copywriting, SEO, email automation, growth strategies, etc.
Equivalent to "modularizing" the capabilities of an entire marketing team.
https://github.com/coreyhaines31/marketingskills

27. Claude SEO
Provides full-site audits, structured data validation, keyword analysis, and other functions.
12 sub-skills in total, covering the complete SEO workflow.
https://github.com/AgriciDaniel/claude-seo

28. Obsidian Skills
Developed by the Obsidian CEO. Supports automatic tagging, automatic linking, and deep integration with the knowledge base (vault).
If you use Obsidian, this is basically a must-install.
https://github.com/kepano/obsidian-skills

29. Context Optimization
Used to reduce token costs and improve KV cache efficiency.
Can significantly reduce API usage costs. GitHub 13,900+ stars.
https://github.com/muratcankoylan/agent-skills-for-context-engineering

30. Deep Research Skill
8-stage research process + automatic continuation mechanism.
Suitable for scenarios where Claude needs to do "deep research" rather than shallow summaries.
https://github.com/199-biotechnologies/claude-deep-research-skill

Part 5: Local AI & Model Running

Run models on local devices: higher privacy, faster response, no API cost.

31. Ollama
Run open-source large models locally with one command. Supports Llama, Mistral, Gemma, etc.
The fastest path from zero to local AI.
https://github.com/ollama/ollama

32. Open WebUI
A self-hosted ChatGPT-like interface. Simple, fast, fully functional.
Used with Ollama, can build a completely private AI system.
https://github.com/open-webui/open-webui

33. LlamaFile
Packages the entire large model into a single executable file.
No dependencies, download and run, extremely simple.
https://github.com/Mozilla-Ocho/llamafile

34. Unsloth
Increases model fine-tuning speed by 2x, while reducing memory usage by 70%.
If you need to train your own model, this is a great starting point.
https://github.com/unslothai/unsloth

35. vLLM
High-throughput inference engine, performance is 2–4x faster than traditional solutions.
Currently one of the mainstream choices for open-source model production deployment.
https://github.com/vllm-project/vllm

Part 6: Workflow & Automation

Integrate AI into your existing tools and processes.

36. n8n
Open-source workflow automation tool, supports 400+ integrations and AI nodes, can be self-hosted.
Currently one of the strongest visual AI automation builders.
https://github.com/n8n-io/n8n

37. Langflow
Build agent workflows by dragging and dropping. Over 140k stars on GitHub.
No need to write code to build complex agent pipelines.
https://github.com/langflow-ai/langflow

38. Huginn
Self-hosted web agents for monitoring, alerts, and data collection.
Emphasizes privacy-first, all automation runs on your own server.
https://极速赛车开奖结果记录

39. DSPy
Uses "programs" rather than "prompts" to drive models.
A framework originating from Stanford research, suitable for scenarios requiring higher stability.
https://github.com/stanfordnlp/dspy

40. Temporal
A "persistent workflow engine" for long-running tasks.
When your automation processes need to handle crashes, retries, timeouts, etc., this is the standard solution.<极速赛车开奖结果记录>

41. GPT Researcher
Automated research agent that can generate structured reports.
Input a topic, output a complete analysis with source citations.
https://github.com/assafelovic/gpt-researcher

42. Firecrawl
Converts any website into a data format suitable for LLMs.
A crawler tool designed specifically for AI pipelines.
https://github.com/mendableai/firecrawl

43. Vanna AI
Natural language to SQL.
Ask questions in English, directly generates database query statements.
Suitable for those who don't want to write SQL but need to use data.
https://github.com/vanna-ai/vanna

44. Instructor
Uses Pydantic models to make any LLM output structured JSON.
Supports 15+ providers including OpenAI, Anthropic, Google, etc.
The practical choice for many production-level AI engineers.
https://python.useinstructor.com

45. Chroma
Open-source vector database.
One of the simplest ways to add semantic search and long-term memory to AI applications.
https://github.com/chroma-core/chroma

46. dlt
Data pipeline tool for LLMs, can connect to 5000+ data sources.
Helps you import various data into AI workflows.
https://github.com/d极速赛车开奖结果记录

47. ExtractThinker
An "ORM" for document intelligence.
Can extract structured data from any type of document.
https://github.com/enoch3712/ExtractThinker

Part 8: API & Infrastructure

The "underlying pipelines" that make everything run in a production environment.

48. FastAPI
Python web framework for deploying AI applications.
Extremely well-documented, with built-in Pydantic data validation.
https://github.com/tiangolo/fastapi

49. Portkey Gateway
Access 250+ LLMs through one API.
Switch models without changing code.
https://github.com/Portkey-AI/gateway

50. OmniRoute
API proxy supporting 44+ AI providers.
Provides load balancing, failover, and cost optimization.
https://github.com/diegosouzapw/OmniRoute

51. lmnr
Used to track and evaluate agent behavior.
Can clearly see what the agent is doing and measure whether its performance meets standards.
https://github.com/lmnr-ai/lmnr

极速赛车开奖结果记录
Turns your codebase into a "persistent knowledge graph".
Allows Claude to remember the entire project structure across sessions.
https://github.com/DeusData/codebase-memory-mcp

Part 9: Curated Collections & Learning Resources

Where to continuously obtain information and constantly iterate cognition.

53. Awesome Claude Skills
Curated collection of skills, over 22k stars on GitHub.
The首选 entry point for finding new skills.
https://github.com/travisvn/awesome-claude-skills

54. Anthropic Skills Repo
Anthropic's official skills repository.
Also the current "standard paradigm" for skill building.
https://github.com/anthropics/skills

55. Awesome Agents
A curated list summarizing 100+ open-source agent tools.
https://github.com/kyrolabs/awesome-agents

56. PromptingGuide
A complete Prompt Engineering guide covering from basics to advanced.
https://www.promptingguide.ai

57. Anthropic Prompt Engineering Tutorial
Includes 9 chapters + Jupyter Notebook hands-on exercises.
One of the best paths to systematically learn Prompting.
https://github.com/anthropics/prompt-eng-interactive-tutorial

58. SkillsMP
A marketplace platform with 80k+ community skills.
Currently the largest Claude skills directory.
https://skillsmp.com

59. MAGI//ARCHIVE
Daily updates of the latest AI project repositories.
Used to track cutting-edge progress.
https://tom-doerr.github.io/repo_posts/

60. Anthropic Official Documentation
Covers all core content including API, Prompt, tool calls, agents, etc.
If you are serious about building AI products,建议 reading this from cover to cover.
https://docs.anthropic.com

How to Actually Use This List

Don't try to install all 60 of these tools at once. That will only lead to information overload and waste your time.

I recommend using it like this:

If you are a developer: Start with Claude Code (01) + Superpowers (05) + Context7 (17) + Tavily (16).
This combination can help you set up a powerful AI programming environment with search capabilities and documentation support.

If you are a creator / knowledge worker: Start with OpenClaw (08) + Obsidian Skills (28) + PDF Processing (23) + Frontend Design (24). This combination gives you an AI assistant with file management, document processing, and content creation capabilities.

If you are building a product: Start with FastAPI (48) + Instructor (44) + Chroma (45) + LangGraph (09).
This combination covers backend framework, structured output, memory system, and agent orchestration capabilities, enough to support a production-level AI application.

If you just want to learn: Start with the Anthropic tutorial (57) + PromptingGuide (56) + Anthropic official documentation (60).
Lay a solid foundation first, then add tools later.

Choose a path and start深入 using it.

Expand your tools slowly as needs arise.

TL;DR

Skills = Teach AI "how to do better"
MCP = Give AI "the ability to access external tools and data"
Repos = Provide all the open-source infrastructure

Combine these three, and what you get is no longer a "demo that looks cool," but a truly productive AI workflow.

That's it. 60 tools. Now, go build something.