MCP App StoreLe meilleur de l'écosystème IA et MCP, sélectionné chaque jour.
Sources
Hugging Face integrates Vision Language Model (VLM) alignment into the TRL library, enabling more robust training and fine-tuning of multimodal models. This simplifies the process for developers to align visual and textual representations in open-source AI.
OpenAI introduces a new 'safe-completions' approach for GPT-5, moving away from rigid hard refusals toward nuanced, output-centric safety training. This aim is to improve both safety and helpfulness when handling dual-use prompts.
The GPT-5 System Card details a unified model routing system that distributes tasks between gpt-5-main, gpt-5-thinking, and gpt-5-thinking-nano. This architecture optimizes for speed and intelligence based on the specific developer use case.
A showcase of how leading developers are utilizing the first iterations of GPT-5. It highlights real-world applications and early performance gains in complex workflows.
OpenAI has officially launched GPT-5, featuring state-of-the-art performance in coding, mathematics, and visual perception. This represents a significant leap in general intelligence and reasoning capabilities.
OpenAI has released GPT OSS, a new family of open-source models available on Hugging Face. This marks a significant shift in their release strategy, providing the developer community with high-performance weights for local deployment and research.
A practical tutorial on using MCP prompts and resource templates to build workflow automation — walking through a meal planning example that demonstrates how prompts can chain resources and tools into reusable, composable workflows. Prompts are an underused MCP primitive and this is one of the clearest guides to their practical value.
The MCP project introduced its formal governance model: Specification Enhancement Proposals (SEPs), a defined maintainer structure, and a community-driven process for evolving the protocol. This post marked MCP's transition from an Anthropic-led project to a community-governed open standard, laying the groundwork for the later Linux Foundation donation.
A guide on building MCP servers using Python to create an AI-powered shopping assistant integrated with Gradio. Demonstrates the practical implementation of the Model Context Protocol for tool-use in AI apps.
Hugging Face releases the `hf` CLI, a streamlined tool designed to make interacting with the HF Hub faster and more intuitive for developers.
A behind-the-scenes look at how Anthropic’s own teams use Claude Code day-to-day: debugging production issues, navigating unfamiliar codebases, building internal automation tools, and accelerating code review. The post doubles as a practical guide with real patterns teams have found most effective. Worth reading for anyone trying to get more from Claude Code.
NVIDIA NIM is accelerating the deployment of LLMs on Hugging Face, providing optimized inference containers for a wide range of models. This integration streamlines the path from model selection to production-ready deployment.
Hugging Face updates its Gradio MCP servers with five significant improvements. This enhances the ability to integrate Gradio-based AI apps into MCP-compatible clients.
A new release of state-of-the-art paired encoders and decoders via the Ettin Suite. Useful for developers working on complex embedding and generation tasks.
Hugging Face is transitioning its Hub from Git LFS to Xet to improve performance and scalability for large datasets. This move aims to optimize how massive AI models and datasets are stored and retrieved.
Kimina-Prover leverages test-time reinforcement learning search to enhance the formal reasoning capabilities of large language models. The research focuses on improving accuracy in complex mathematical and logical proofs.
Research on improving robotic efficiency by decoupling action prediction from execution. Relevant for those building high-performance agentic physical systems.
Introduction of ScreenEnv, a framework for deploying full-stack desktop agents. Provides the necessary environment for agents to interact with OS-level interfaces.
Hugging Face introduces a new MCP server, expanding the Model Context Protocol ecosystem by allowing AI agents to interact more deeply with the HF Hub. This enables better discovery and integration of open-source models and datasets within agentic workflows.
Hugging Face introduces Gradio MCP Servers, allowing LLMs to interact directly with Gradio apps. This expands the MCP ecosystem by bridging the gap between interactive ML demos and agentic tool-use.