scrubby.ai

What it is

Scrubby is an AI code review and repository intelligence platform for engineering teams. It is not primarily positioned as a replacement for Claude Code, Cursor, or Copilot when writing code. Instead, it adds a “understand this repository” context layer for these AI agents: automatically learning the codebase’s architectural domains, module boundaries, team conventions, naming habits, and co-change relationships from Git history, helping reduce the problem of AI-generated code that “runs, but doesn’t look like code this team would write.”

Core capabilities and integrations

Scrubby’s analysis is organized into three layers: Map, Understand, and Learn. It discovers codebase domains, tracks imports, file clustering, module boundaries, and coupling, then mines Git history for co-changes, common patterns, and conventions. It connects to tools such as Claude Code, Cursor, VS Code, Windsurf, and Zed through a remote HTTP MCP Server, and also provides a GitHub App for automated PR review. Available tools include file-level review, changeset review, domains/network queries, and findings feedback, making it useful for providing structured context before editing, before committing, and during the PR stage.

Pricing and privacy

At the moment, the official website only offers Join the Beta / Request access, with no disclosed plans, free quota, enterprise pricing, or payment methods. On privacy, Scrubby says it does not persist full source code, but stores file metadata, AI summaries, and architecture metrics. During indexing, trimmed content is sent to its servers and processed via the Anthropic API, then discarded after summaries are generated. This explanation is more specific than many early-stage tools, but enterprise procurement would still need further assessment of data residency, compliance certifications, DPA availability, and self-hosting or private deployment options.

Pros, cons, and limitations

Its main strength is a very clear positioning: unlike traditional linters, it learns each repository’s own rules; unlike built-in editor search, it focuses on domains, architecture, and historical co-change patterns, and is not tied to a single IDE. It explicitly states support for monorepos, JS/TS, Python, Ruby, Go, Java, and frameworks such as React, Next.js, Rails, and Django. The limitations are that the product is still in Beta, pricing and support levels are unknown, real-world accuracy has not been independently validated, and results depend on the quality of the repository’s history, code structure, and whether the AI agent consistently calls MCP tools.

Who it’s for and access from China

Scrubby is a good fit for teams that make heavy use of AI coding agents, have large codebases, high PR review costs, or knowledge silos. It can also help open-source maintainers reduce the onboarding cost for contributors. For small personal projects without complex architecture or multi-person collaboration, the benefits may be limited. The official site does not specify accessibility from China; because it depends on GitHub OAuth, a remote MCP service, and the Anthropic API, actual network connectivity and account availability may vary by environment. Alternatives to watch include CodeRabbit, Codacy, GitHub Copilot/Code Review, Sourcegraph Cody, and Cursor’s built-in codebase indexing.