MCP App StoreLe meilleur de l'écosystème IA et MCP, sélectionné chaque jour.
Yesterday was dominated by advancements in agent steering and enterprise-grade deployment for MCP, alongside a critical look at the security vulnerabilities inherent in agentic workflows.
Anthropic significantly expanded the control surface for Claude Code, introducing a robust steering framework via CLAUDE.md and subagents to make AI-driven coding more predictable. Simultaneously, the Model Context Protocol (MCP) moved closer to enterprise readiness with zero-touch OAuth, removing the friction of manual server authentication.
However, the "MosaicLeaks" research serves as a timely warning for developers: as agents aggregate more data from diverse sources, the risk of inadvertent sensitive data leakage increases.
Today's stories:
The day reflects a clear tension between increasing agent autonomy and the urgent need for tighter security and control frameworks.
ServiceNow explores the vulnerability of research agents to data leakage through 'MosaicLeaks'. The study examines how agentic workflows can inadvertently expose sensitive information when aggregating data from multiple sources.
The Model Context Protocol now supports Enterprise-Managed Authorization, allowing organizations to centrally provision server access via identity providers. This enables zero-touch OAuth, connecting users to servers on first login without manual per-app authentication.
Claude Code now features live, interactive Artifacts that allow developers to preview in-progress work directly from the session context. These artifacts are shareable with teams, bridging the gap between CLI coding and visual feedback.
Anthropic detailed seven primary methods for controlling Claude Code's behavior, including CLAUDE.md files, skills, and subagents. This framework allows developers to manage the authority and context cost of instructions for more predictable agentic behavior.
Hugging Face analyzes alternatives to Low-Rank Adaptation (LoRA) for efficient model fine-tuning. The piece evaluates whether newer PEFT techniques can outperform LoRA in terms of performance and resource efficiency.