AI News Brief: June 6, 2026

Claude Code Introduces Multi-Repository Context Window

Anthropic has expanded Claude Code's context handling to simultaneously track dependencies across up to 10 linked repositories. The update enables developers to refactor code that spans multiple services without manual context switching. Early adopters report 40% faster completion times for microservices architecture work.

MCP Gateway Protocol Reaches v2.0 Milestone

The Model Context Protocol consortium released MCP v2.0, introducing standardized streaming interfaces for real-time data sources and a new authentication layer for enterprise deployments. Over 200 tools and data connectors now support the updated specification. The release includes backwards compatibility adapters for v1.x implementations.

OpenHands Agent Framework Adds Visual UI Testing

The open-source OpenHands project launched autonomous browser testing capabilities that allow AI agents to validate web interfaces end-to-end. Agents can now generate test cases, execute them across multiple browsers, and file detailed bug reports with screenshots. The feature integrates with Playwright and Selenium frameworks.

Cursor Editor Debuts Collaborative AI Pair Programming

Cursor released a beta feature enabling multiple developers to share an AI coding session with persistent context. The shared workspace allows team members to ask follow-up questions on each other's AI-generated code and maintains a unified conversation history. Premium tier users gain access starting next week.

Llama-Coder 4.0 Achieves 89% on HumanEval Benchmark

Meta's latest open-source coding model demonstrates significant improvements in code generation accuracy, matching some proprietary alternatives. The 70B parameter model runs efficiently on consumer hardware with 4-bit quantization. Community fine-tunes for Rust, Go, and TypeScript are already available on HuggingFace.