AI News for June 25, 2026

Claude Code Launches Multi-Repository Context Management

Anthropic released a major update to Claude Code enabling seamless context sharing across up to 50 linked repositories simultaneously. The feature uses an improved MCP server architecture that maintains semantic understanding of codebases up to 10M lines while keeping latency under 200ms. Early adopters report 40% faster refactoring workflows in polyrepo environments.

MCP 2.1 Specification Adds Streaming Tool Execution

The Model Context Protocol consortium published version 2.1, introducing native support for streaming tool outputs and partial result handling. This enables AI agents to process long-running operations like database migrations or ML training jobs without timeout issues. Over 30 MCP-compatible tools have already committed to supporting the new streaming primitives.

Open-Source Agent Framework "Cascade" Reaches 1.0

The collaborative agent orchestration framework Cascade hit its first stable release after 18 months of development. Built on a composable task graph architecture, it allows developers to chain multiple AI models and tools with automatic error recovery and state management. The project has gained 28K GitHub stars and includes pre-built integrations for all major LLM providers.

Cursor Introduces "Time-Travel Debugging" for AI-Generated Code

Cursor IDE shipped a novel debugging feature that creates automatic checkpoints before each AI code generation, allowing developers to instantly revert or compare previous iterations. The system uses lightweight AST snapshots and integrates with Git to provide seamless version control without repository pollution. The feature is available in the latest Cursor 0.47 release.