AI News Brief: April 1, 2026

Claude Code Introduces Multi-Repository Context Window

Anthropic expanded Claude Code's capabilities to maintain context across up to 10 linked repositories simultaneously, with a combined context window of 500K tokens. The feature uses intelligent caching to prioritize frequently accessed code paths and automatically maps dependencies between projects. Early beta testers report 40% faster resolution of cross-service bugs.

MCP 2.0 Specification Released with Bidirectional Streaming

The Model Context Protocol consortium published version 2.0, introducing bidirectional streaming that allows AI agents to push real-time updates to connected tools and services. The update includes standardized authentication flows and a new resource subscription model that reduces polling overhead by 85%. Over 200 tools have committed to supporting the new spec within 60 days.

Open-Source Agent Framework "Nexus" Hits 1.0

The Nexus framework reached production readiness with built-in support for multi-agent collaboration, persistent memory, and automatic tool discovery via MCP. Written in Rust with Python bindings, it features a novel "consensus layer" where multiple AI models vote on high-stakes decisions. The project has gained 15K GitHub stars since its beta launch in February.

GitHub Copilot Workspace Adds Visual Debugging Mode

GitHub introduced a visual debugging interface for Copilot Workspace that renders execution flows as interactive diagrams. Developers can click on any node to see AI-generated explanations of variable states and suggest fixes for runtime errors. The feature integrates with popular IDEs and supports 12 programming languages at launch.