AI News Summary - March 4, 2026

GitHub Copilot Workspace Adds Multi-Repository Context

GitHub announced that Copilot Workspace now supports cross-repository context awareness, allowing developers to work on microservices architectures with shared context across up to 10 related repositories. The update uses an improved indexing system that maintains semantic relationships between codebases while staying within token limits. Early access users report 40% faster feature implementation when working across distributed systems.

Anthropic Releases MCP Server Registry with 500+ Integrations

Anthropic launched an official Model Context Protocol server registry, featuring over 500 community-contributed and verified MCP servers for databases, APIs, and development tools. The registry includes automated security scanning, version management, and one-click installation for Claude Desktop and compatible applications. Popular additions include servers for Kubernetes management, Terraform state inspection, and real-time log analysis.

Open-Source Agent Framework "Conductor" Reaches 1.0

The Conductor project, an open-source framework for building multi-agent systems with built-in observability, released version 1.0 after 18 months of development. It features declarative agent composition, automatic failover between LLM providers, and native support for tool calling with OpenAI, Anthropic, and local models. The framework has already been adopted by several enterprises for customer support automation and internal tooling.

Cursor Introduces "Shadow Mode" for Safe AI Code Generation

Cursor rolled out Shadow Mode, a feature that runs AI code suggestions in an isolated sandbox environment before applying them to actual codebases. The system automatically tests generated code against existing unit tests and performs static analysis to catch potential bugs or security issues. Users can review a diff showing what would have happened, reducing the risk of AI-generated errors in production code.