AI News Summary - October 12, 2026

Claude Code Introduces Multi-Repository Context Window

Anthropic announced that Claude Code can now maintain context across up to 10 linked repositories simultaneously, with a combined 500K token context window. This enables developers to refactor code across microservices architectures while maintaining consistency. The feature is available in Claude Code Pro starting today.

MCP 2.0 Specification Released with Bidirectional Streaming

The Model Context Protocol consortium published MCP 2.0, introducing bidirectional streaming capabilities that allow AI agents to maintain persistent connections with external tools and data sources. Major adopters including Microsoft, Google, and Anthropic have committed to implementation timelines within Q4 2026. The new spec reduces latency by 60% for tool-heavy agent workflows.

Cursor Launches "Team Memory" for Collaborative AI Coding

Cursor unveiled Team Memory, a workspace feature that learns from collective team coding patterns and maintains shared context across all team members' AI interactions. The system adapts suggestions based on team-specific conventions, approved libraries, and architectural decisions. Early access teams report 40% fewer code review iterations.

Open-Source AgentKit Framework Reaches 1.0

AgentKit, the open-source framework for building production-grade AI agents, hit version 1.0 with built-in support for task decomposition, error recovery, and human-in-the-loop workflows. The framework now includes pre-built integrations for 50+ tools and services, with over 15,000 GitHub stars since its beta launch six months ago.