MCP App StoreThe latest from the AI and MCP ecosystem, curated daily.
This week marked a decisive pivot from LLMs as "chatbots" toward autonomous, specialized agentic systems. The focus shifted from raw model size to architectural efficiency, orchestration logic, and physical-world interaction.
Anthropic dominated the narrative with a blueprint for AI-native engineering. The introduction of dynamic multi-agent harnesses in Claude Code and the release of the Claude Cowork product guide signal a move toward modular orchestration where specialized temporary agent structures handle complex professional workflows. This is further supported by IBM Research, arguing that "agent laogic" is the fundamental requirement for scalable enterprise AI reliability.
The boundary between high-performance AI and local hardware continues to blur. Google released Gemma 4 12B and la accompanying QAT checkpoints, optimizing multimodal intelligence for laptops and mobile devices. Simultaneously, Hugging Face demonstrated the viability of multi-agent economies running on models as small as 3B, proving that emergent agentic behavior is no longer tethered to frontier-scale compute.
NVIDIA's Cosmos 3 omni-model and the integration of MCP tools into the Reachy Mini robot highlight a growing trend: the extension of agentic reasoning into physical action. Whether it is robotics or the "computer use" latency optimizations seen in Holo3.1, the goal is now real-time, reliable interaction with the external world.
Key Stories:
Exploration of a multi-agent economy simulated using a small 3B parameter model. Demonstrates the viability of complex agent interactions and emergent behavior in highly constrained model sizes.
Google releases Quantization-Aware Training (QAT) checkpoints for Gemma 4. These optimizations reduce memory overhead and significantly improve performance for on-device deployment on laptops and mobile devices.
Cursor introduces Design Mode, allowing developers to point, draw, or narrate UI changes directly in the browser. AI agents then automatically update the underlying code to match the visual requests.
.jpg)
A case study on how Claude Code is used internally at Anthropic to automate GTM engineering tasks. It demonstrates significant time savings on repetitive research and email workflows.
Anthropic releases the official product guide for Claude Cowork. The guide covers initial setup and best practices for deploying AI agents to handle complex professional tasks.
NVIDIA releases Nemotron 3.5 Content Safety, providing customizable multimodal safety guardrails for enterprise AI. It focuses on scaling safety checks across global deployments and various data modalities.
Google has launched local development capabilities for Kaggle Benchmarks, allowing developers to create and test AI benchmarks more efficiently. This streamlines the evaluation process for AI models on local machines.
ServiceNow AI introduces EVA-Bench Data 2.0, a comprehensive dataset for evaluating agentic AI across 121 tools and 213 scenarios. This provides a robust framework for measuring tool-use and task-completion capabilities of AI agents.
NVIDIA introduces a new method for generating high-quality synthetic Q&A pairs to improve the pretraining of Nemotron models. This approach leverages task-seeded generation to create more diverse and accurate training data.
OpenAI introduces a new memory system for ChatGPT designed to better remember user preferences and maintain context across conversations. This improvement aims to make the assistant more personalized and relevant over the long term.
Hugging Face introduces a redesigned CLI optimized for AI agents to interact with the Hub. This enables more seamless model and dataset management within agentic workflows.
Google releases Gemma 4 12B, a high-performance multimodal model designed for local execution on laptops. It features a unified, encoder-free architecture to bring advanced intelligence to the edge.
OpenAI enhances GPT-Rosalind with advanced biological reasoning, genomics analysis, and medicinal chemistry expertise. This update aims to accelerate life sciences research and experimental workflows.
Dharma-AI explores the application of Direct Preference Optimization (DPO) outside of traditional chatbot contexts. This research expands the utility of preference alignment for broader AI tasks.
Wasmer leveraged Codex with GPT-5.5 to develop a Node.js runtime for the edge, drastically reducing development time. The project demonstrates the power of frontier models in systems engineering.
A practical implementation of the Model Context Protocol (MCP) to add tools to the Reachy Mini robot. This showcases how MCP can bridge LLMs with physical robotic hardware.
Anthropic provides a guide on effectively using Claude Cowork compared to Claude Code or standard chat. It focuses on workflow delegation and concrete steps for integrating agentic collaboration into development.
A deep dive into how Anthropic built and scaled hundreds of internal skills for Claude Code. It provides valuable lessons on agentic tooling and the architecture of specialized capabilities.
Anthropic shares strategies for using Claude to drive self-service data insights. The post highlights practical approaches to maximizing the model's analytical capabilities for data-driven decision making.
Cursor introduces Organizations, allowing enterprises to manage multiple teams with centralized control over budgets, security, and governance. This simplifies scaling AI-powered development across large companies.
Anthropic shares how the Claude Code team restructured their engineering processes to make agentic coding the default workflow. It provides a blueprint for scaling AI-native development practices in a professional engineering organization.
Hcompany introduces Holo3.1, focusing on high-performance local computer use agents. This release emphasizes reduced latency and improved local execution for agentic UI interactions.
OpenAI reports on how Codex is evolving into a broad productivity tool for knowledge work, automating research and data analysis workflows. This signals a shift toward more integrated AI-driven professional productivity.
Claude Code now supports dynamic multi-agent harnesses, allowing it to orchestrate complex workflows on the fly. This enables more robust task execution through specialized, temporary agent structures.
JetBrains has released Mellum2, a 12B parameter Mixture-of-Experts (MoE) model. The model is designed for high-performance AI tasks, continuing JetBrains' push into specialized developer-centric AI models.
IBM Research explores the necessity of 'agent logic' for scaling AI in enterprises. The piece argues that moving beyond simple LLM prompts to structured agentic workflows is key to reliable AI adoption.
OpenAI frontier models and Codex are now generally available on AWS. This integration allows enterprises to leverage OpenAI's capabilities within their existing AWS environments and procurement workflows.
NVIDIA has released Cosmos 3, an open omni-model designed for physical AI reasoning and action. This model aims to bridge the gap between digital intelligence and physical world interaction, providing a foundation for robotics and autonomous systems.