MCP App StoreLe meilleur de l'écosystème IA et MCP, sélectionné chaque jour.
Sources
OpenAI shares initial research on GPT-5's ability to accelerate scientific discovery in math, physics, and biology. This signals a shift toward high-reasoning capabilities specifically tuned for complex scientific proof and insight generation.
Explores the distillation of reasoning capabilities into more efficient models via Apriel-H1. This research provides insights into creating high-performance reasoning models with lower compute overhead.
OpenAI introduces GPT-5.1-Codex-Max, a high-performance agentic coding model designed for large-scale project work. It features improved reasoning and token efficiency for long-running development tasks.
OpenAI released the system card for GPT-5.1-Codex-Max, detailing safety mitigations for harmful tasks and prompt injections. Key technical highlights include agent sandboxing and configurable network access to secure model execution.
Hugging Face introduces a streamlined workflow for building and sharing ROCm kernels, simplifying the optimization of AI models on AMD hardware. This is a significant win for open-source accessibility on non-NVIDIA GPUs.
OpenAI explores mechanistic interpretability using a sparse model approach to decode how neural networks reason. This research aims to increase transparency in AI decision-making and improve system reliability and safety.
OpenAI introduces group chats in ChatGPT, allowing multiple users to collaborate with the AI in a single shared conversation. This enhances team productivity and real-time AI-assisted brainstorming.
OpenAI releases GPT-5.1 with enhanced adaptive reasoning, improved coding performance, and extended prompt caching. New apply_patch and shell tools provide deeper agentic control for developers.
.jpg)
Anthropic shares best practices for using Agent Skills to build richer, more customised frontend UIs with Claude. The post covers how skills can encode design conventions, component libraries, and project-specific patterns so Claude consistently produces on-brand output. Directly relevant to teams using Claude Code or Skills for frontend development work.
OpenAI explores the mechanics of prompt injection attacks and the research being conducted to build more robust safeguards. This is critical for developers building agentic workflows where user input can bypass system prompts.
Anthropic demonstrates how combining MCP with code execution lets agents handle far more tools while consuming dramatically fewer tokens — up to 98.7% reduction in context overhead in their tests. Instead of passing tool results back through the context window, agents write code to chain tool calls and filter outputs before they reach the model. A major efficiency win for complex multi-tool agent pipelines.
An underused MCP protocol feature gets a spotlight: server instructions let you give the LLM a structured user manual for your server — explaining what tools do, when to use them, and how to combine them effectively. The post shows how well-written server instructions dramatically improve tool call accuracy without any prompt engineering on the client side.
OpenAI has unveiled Aardvark, an autonomous AI security researcher designed to identify and validate software vulnerabilities at scale. This represents a significant step toward agentic security tooling for developers and security teams.
MiniMax explores agent generalization in their M2 model, questioning how agents align to goals across diverse tasks. This research provides insights into improving the reliability and adaptability of agentic AI systems.
Anthropic covers practical patterns for deploying Claude-based agents in financial services: document processing, compliance checks, customer support automation, and multi-step research workflows. The post addresses the specific trust and auditability requirements that make fintech agent deployments more complex than typical use cases. Useful for teams building regulated-industry agents.
OpenAI details OWL, the underlying architecture for the Atlas browser. By decoupling Chromium, it enables faster startup and provides a foundation for deeper agentic browsing capabilities within ChatGPT.
OpenAI released gpt-oss-safeguard, a set of open-weight reasoning models designed for safety classification. These models allow developers to implement and iterate on custom safety policies for their own AI applications.
Dagster Labs shares how they use Codex across their open-source projects to accelerate documentation — generating docs from code, translating content across formats, and measuring doc completeness automatically. A practical case study showing Codex applied to developer tooling maintenance rather than greenfield code generation.
OpenAI has acquired Software Applications Incorporated to integrate Sky, a natural language interface for macOS, into ChatGPT. This move aims to enhance ChatGPT's desktop experience with deep system-level capabilities, making it more action-oriented and contextual.
Consensus has launched a multi-agent research assistant powered by GPT-5 and the OpenAI Responses API. The tool enables researchers to read, analyze, and synthesize scientific evidence in minutes, significantly accelerating discovery workflows.