Microsoft open-sources Scope, a testing tool for AI coding agents

Microsoft has released Scope on GitHub under the MIT license; it runs the same task through different coding agents such as GitHub Copilot and Claude Code and scores each run against criteria forming a dependency graph. An LLM-based Judge service grades results, while logs, workspace snapshots and tool calls are kept for tracing failures. There is no hosted version: local setup needs Docker Compose v2, Node 22, pnpm, mkcert and the GitHub CLI, or a Kubernetes deployment via Helm chart.

• Scope is released under the MIT license, code on GitHub
• Success is checked against dependency-graph criteria scored by an LLM judge
• Local setup requires Docker Compose v2, Node 22, pnpm, mkcert and GitHub CLI
• CLI, REST API and a Helm chart support shared Kubernetes deployments