thedaviddias/Front-End-Checklist
Front-End Checklist is an open-source quality system for modern web development, offering comprehensive best practices for humans and AI agents.
Awesome AI Agents › MCP Server Framework
HKUDS/AnyTool
AnyTool from HKUDS is a universal tool-use layer that sits between an AI agent and the tools it can call. It targets three problems the authors identify with current agent tooling: contexts that become overwhelming once many tools are registered, community-published tools of uncertain reliability, and coverage limited to whatever a single integration supports. The retrieval path uses a multi-stage progressive filtering pipeline that narrows a large catalogue to the few tools relevant to the current request in milliseconds, so the prompt is not polluted with unused definitions. Embeddings are precomputed and initialisation is lazy, which the project describes as removing redundant work between executions. For scale, adaptive Model Context Protocol tool selection combines caching with selective re-indexing, with the stated aim of holding performance roughly constant from ten tools up to ten thousand as the ecosystem grows. A persistent memory records how tools have behaved, and this reliability tracking feeds a quality-aware selection step alongside execution safeguards, so poor or unsafe tools are avoided over time rather than retried. The backend architecture is multi-modal in the operational sense: beyond web APIs it can drive shell and system operations, GUI automation and deep-research flows through one unified interface, and the shell and GUI backends can run either locally or against a server. Integration is deliberately small: importing AnyTool and using it as an async context manager, then passing a natural-language instruction to execute, runs the whole orchestration, as shown by an example that researches AI coding tools across GitHub and news sources and produces a comparison report. Setup uses a conda environment on Python 3.12 with ffmpeg for video recording, dependencies from requirements.txt, and API keys in a .env file. It runs on macOS, Linux and Windows under the MIT licence.
https://github.com/HKUDS/AnyTool
Also in MCP Server Framework
PrefectHQ/fastmcp
FastMCP offers a Pythonic framework for building Model Context Protocol (MCP) servers and clients, facilitating structured communication for AI agents and models.
pascalorg/editor
Local-first 3D building editor built with React Three Fiber and WebGPU that exposes an MCP service so AI agents can create and edit 3D scenes.
trycua/cua
c/ua is a high-performance framework that enables AI agents to control full macOS and Linux operating systems within secure, lightweight virtual containers on Apple Silicon Macs, supporting complex autonomous workflows and LLM integration.
nukeop/nuclear
Nuclear is a free, open-source, ad-free music player for Windows, macOS, and Linux, featuring streaming from various sources, playlist management, a powerful plugin system, and integration with AI ...
googleapis/mcp-toolbox
MCP Toolbox for Databases is an open-source server that connects AI agents and applications to enterprise databases using the Model Context Protocol (MCP).
apify/apify-mcp-server
Model Context Protocol server that exposes thousands of Apify Actors as tools, letting AI agents scrape social media, search engines, maps and e-commerce sites over hosted HTTP or local stdio.
grab/cursor-talk-to-figma-mcp
Cursor Talk to Figma MCP is an integration that enables Cursor AI to communicate with Figma programmatically for reading and modifying designs, automating and enhancing the design workflow.