MCP Tools
Complete reference for the Scrubby MCP tools your AI editor can call. One page per tool with parameters, response shape, and examples.
The Scrubby MCP server exposes eight tools. Once connected, your AI editor (Claude Code, Cursor, Windsurf, VS Code, Zed) can call any of them in the same way it calls read_file or grep — the server publishes tool descriptions over the protocol, and the agent picks the right one for the task.
This index summarizes each tool and links to a dedicated reference page.
Indexing
scrubby_index— Index a repository for the first time, or re-index after major changes.
Codebase intelligence (read)
scrubby_review— Codebase knowledge for a specific file: domain, conventions, connected files, git history.scrubby_review_changeset— Review a set of changed files together: missing co-changes, consistency issues, domain crossings.scrubby_get_domains— List all discovered domains with patterns, file counts, languages.scrubby_get_segments— List all code segments with their conventions and file summaries.scrubby_get_network— Domain connection graph with Hebbian-learned weights.scrubby_get_findings— Commit and domain activity history for the repository.
Learning loop (write)
scrubby_report_findings— Report review findings so Scrubby can refine its connection weights.
Common parameters
Most read tools accept the same two optional disambiguation parameters:
repo_id— Scrubby repository ID. Auto-detected if omitted.repo_name— Repository name (e.g.api). Used to auto-resolve in remote mode.
Authentication
All tools require an authenticated session. The first time your editor connects, you’ll be prompted through GitHub OAuth. After that, Scrubby keeps the session alive automatically. If you see “not authenticated” errors, see Auth & OAuth Issues.
Versioning and stability
The MCP tool surface is stable in name and shape. Additive changes (new optional parameters, new fields in the response) are routine. Breaking changes are announced in advance. If you’re scripting against the MCP tools, prefer the structured response fields documented on each per-tool page over parsing free-text output.
Last updated