AI News Summary - September 2, 2026

Claude Code Introduces Multi-Repository Context Windows

Anthropic announced that Claude Code now supports simultaneous context from up to 10 linked repositories, enabling developers to work across microservices architectures seamlessly. The feature uses an enhanced version of MCP to maintain consistent state across codebases while staying within the 500K token context limit. Early beta testers report 40% faster development velocity on distributed systems projects.

MCP 3.0 Specification Released with Bidirectional Streaming

The Model Context Protocol consortium published MCP 3.0, introducing native support for bidirectional streaming between AI agents and external tools. This enables real-time collaborative editing scenarios and continuous data synchronization without polling overhead. Major adopters including GitHub, JetBrains, and Replit have committed to implementation within Q4 2026.

Open-Source Agent Framework "Nexus" Reaches 1.0

The Nexus agent framework hit production-ready status with built-in support for tool composition, memory persistence, and multi-modal reasoning. Unlike previous frameworks, Nexus uses a declarative YAML configuration for defining agent behaviors and automatically generates MCP-compliant interfaces. The project has gained 45K GitHub stars in three months and already powers several commercial AI applications.

Cursor Launches "Context Maps" for Visual Code Navigation

Cursor's latest update introduces Context Maps, an interactive visualization showing how AI assistants understand project structure and dependencies. Developers can now see which files and functions the AI considers relevant for any given task and manually adjust attention weights. The feature addresses a common complaint about AI coding tools making unexpected changes in seemingly unrelated files.