AI News Brief: March 30, 2026

Claude Code Launches Multi-Repository Analysis Feature

Anthropic's Claude Code now supports analyzing and refactoring across up to 10 linked repositories simultaneously. The update includes intelligent dependency tracking that automatically identifies breaking changes across microservices architectures. Early adopters report 40% faster cross-repo refactoring workflows compared to single-repository operations.

MCP 2.1 Specification Introduces Streaming Context Windows

The Model Context Protocol consortium released version 2.1, featuring streaming context windows that allow AI agents to process real-time data feeds without full context reloads. This enables persistent connections to databases, file watchers, and API streams while maintaining sub-200ms latency. Over 30 tools including Continue and Cody have already committed to supporting the new spec by Q2.

Open-Source Agent-Flow Framework Reaches Production Stability

Agent-Flow, an MIT-licensed orchestration framework for multi-agent systems, hit version 1.0 after 18 months of development. The framework provides visual workflow builders, built-in error recovery, and native MCP support for coordinating specialized AI agents. Initial benchmarks show it handles complex 50+ agent workflows with 99.7% task completion rates.

GitHub Copilot Workspace Adds Natural Language Project Scaffolding

GitHub announced Copilot Workspace can now generate complete project structures from conversational descriptions, including CI/CD pipelines, Docker configurations, and testing frameworks. The feature supports 15 programming languages and automatically applies organization-specific best practices from existing repositories. Beta access opens to GitHub Enterprise customers next week.