Trending GitHub AI Projects & Agentic Tools
Curated breakout open-source repositories for AI workflows. Simple explanations, why they matter, and instant copy-paste configs to connect them with Claude Desktop, Cursor, and AI agents.
REA (Reverse Engineer Anything)
morluto/reaModel Context Protocol (MCP) server bridging AI coding agents (Claude Code, Cursor, Codex) with reverse-engineering engines to inspect binaries, APKs, Electron apps, and web frontends.
Empowers AI agents to understand proprietary binaries, trace execution paths, recover architectural graphs, and debug compiled applications without source code.
Add to your Claude Desktop config (claude_desktop_config.json) or Cursor Settings → MCP.
{
"mcpServers": {
"rea": {
"command": "npx",
"args": [
"-y",
"rea-agents",
"mcp"
]
}
}
} Browser-Use
browser-use/browser-useOpen-source web automation library and agent enabling LLMs to natively control web browsers, fill forms, navigate pages, and extract data.
Translates plain English instructions into real browser actions — clicking elements, handling auth, and executing complex workflows autonomously.
Install via pip and pass natural language tasks to the Agent object with your preferred LLM.
from langchain_openai import ChatOpenAI
from browser_use import Agent
import asyncio
async def main():
agent = Agent(
task="Найди на GitHub трендовые репозитории по MCP и сохрани в файл",
llm=ChatOpenAI(model="gpt-4o"),
)
await agent.run()
asyncio.run(main()) Crawl4AI
unclecode/crawl4aiOpen-source, LLM-friendly web crawler and scraper designed to feed clean, structured Markdown into RAG pipelines and AI agents.
Eliminates HTML boilerplate and navigation menus, delivering pristine Markdown to minimize token usage in vector databases and agent context windows.
Run locally in Python or deploy as an MCP server to supply live web content directly to your coding assistant.
import asyncio
from crawl4ai import AsyncWebCrawler
async def main():
async with AsyncWebCrawler(verbose=True) as crawler:
result = await crawler.arun(url="https://news.ycombinator.com")
print(result.markdown[:1000])
asyncio.run(main()) Ollama
ollama/ollamaGet up and running with large language models locally. Run Llama 3, DeepSeek-R1, Mistral, and custom models in one command.
100% private and offline inference on consumer GPUs/CPUs without telemetry or subscription fees.
Download from ollama.com or run the CLI command to start chatting immediately.
ollama run deepseek-r1:8b Official MCP Reference Servers
modelcontextprotocol/serversOfficial reference implementation of Model Context Protocol servers by Anthropic: Filesystem, GitHub, Postgres, SQLite, Slack, and Fetch.
Equips coding assistants with standardized access to filesystems, databases, and third-party developer APIs out of the box.
Paste snippet into your claude_desktop_config.json or Cursor/Windsurf MCP configuration settings.
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"C:/Projects"
]
},
"fetch": {
"command": "uvx",
"args": [
"mcp-server-fetch"
]
}
}
} FastMCP
jlowin/fastmcpThe fastest, cleanest Python framework for building Model Context Protocol (MCP) servers using simple function decorators.
Transforms Python functions into fully documented MCP tools with automatic schema validation and type hint inference.
Save as server.py and run via `fastmcp dev server.py` or inspect with the MCP inspector.
from fastmcp import FastMCP
mcp = FastMCP("MyAssistantTools")
@mcp.tool()
def calculate_vat(amount: float, rate: float = 22.0) -> float:
"""Вычисляет сумму НДС по заданной ставке."""
return round(amount * (rate / 100), 2)
if __name__ == "__main__":
mcp.run() Open WebUI
open-webui/open-webuiExtensible, feature-rich, and self-hosted WebUI for LLMs supporting Ollama, OpenAI-compatible APIs, document RAG, and multi-user chat.
Turnkey self-hosted ChatGPT replacement with document Q&A, role-based access control, and seamless Ollama integration.
Launch the Docker container and visit http://localhost:3000 in your browser.
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main Mem0 (The Memory Layer for AI)
mem0ai/mem0Universal memory layer for AI agents and assistants. Stores user preferences, session history, and semantic facts across interactions.
Provides persistent personalization so agents remember developer habits, stack constraints, and domain preferences across turns.
Import into your agent workflow and call memory.add() / memory.search() to ground prompts.
from mem0 import Memory
m = Memory()
# Сохранить факт о пользователе
m.add("Пользователь предпочитает Astro 5 и Tailwind CSS", user_id="rashid")
# Получить релевантный контекст перед ответом
memories = m.get_all(user_id="rashid")
print(memories) OpenHands (ex-OpenDevin)
All-Hands-AI/OpenHandsAutonomous AI software development agent capable of writing code, running bash commands, executing tests, and debugging in sandboxed containers.
Acts as an autonomous pair programmer that executes real commands inside an isolated Docker sandbox.
Run the official Docker container and navigate to http://localhost:3000.
docker run -it --rm -p 3000:3000 -v /var/run/docker.sock:/var/run/docker.sock ghcr.io/all-hands-ai/openhands:latest Cline (Autonomous VS Code Agent)
cline/clineAutonomous coding agent for VS Code with terminal execution, browser testing, and custom MCP server tool integration.
Enables safe, human-in-the-loop autonomous programming right inside your existing editor and workspace.
Install the Cline extension in VS Code via the marketplace or the CLI command.
code --install-extension saoudrizwan.claude-dev Dify
langgenius/difyOpen-source LLM app development platform combining agentic workflows, visual orchestration, RAG pipelines, and model management.
Build production-grade AI agents and RAG assistants using an intuitive visual flowchart interface.
Clone the repository and launch the full stack via Docker Compose.
git clone https://github.com/langgenius/dify.git && cd dify/docker && docker compose up -d vLLM
vllm-project/vllmHigh-throughput, low-latency LLM serving engine featuring PagedAttention for production model deployments.
Production standard for serving open-weights LLMs with massive concurrency and OpenAI API compatibility.
Run the CLI command to launch a local server compatible with OpenAI endpoints.
pip install vllm && python -m vllm.entrypoints.openai.api_server --model Qwen/Qwen2.5-Coder-7B-Instruct Composio
ComposioHQ/composioProduction tooling platform providing 250+ pre-built integrations (GitHub, Slack, Google, Jira) for AI agents and LLMs.
Manages user auth, OAuth2 flows, and permissions so AI agents can reliably act on external SaaS tools.
Install the Python SDK and execute `composio add <tool>` to equip agents with integrations.
pip install composio-core Agent Skills (skills.sh)
anthropics/skillsOpen catalog and specification for agent skills (`SKILL.md`) enabling plug-and-play capability extensions for coding agents.
Equips coding assistants with specialized domain expertise, best practices, and repeatable workflows via `npx skills add`.
Use `npx skills add <owner/repo@skill> -y --copy` to equip agents with pre-built skills.
npx skills find react RAGFlow
infiniflow/ragflowNext-generation open-source RAG engine based on deep document understanding of complex PDFs, tables, and multi-format docs.
Preserves document layout, complex tables, and visual structure to eliminate hallucinations in technical document Q&A.
Deploy via the official docker-compose stack and upload enterprise manuals or documentation.
docker compose -f docker-compose.yml up -d Agno (ex-Phidata)
agno-agi/agnoLightweight, ultra-fast Python framework for building multi-modal agents with memory, knowledge, and tools.
Provides clean, minimal agent primitives without bloat or excessive layers of abstraction.
Install via pip and build multi-modal agents with minimal lines of clean Python code.
from agno.agent import Agent
from agno.models.openai import OpenAIChat
agent = Agent(
model=OpenAIChat(id="gpt-4o"),
description="Вы ведущий системный инженер",
markdown=True
)
agent.print_response("Объясни устройство протокола MCP за 3 предложения.")