Understand-Anything

TL;DR

• Claude Code skill (slash command /understand) that orchestrates a 5-agent pipeline to parse a codebase into an interactive structural and domain knowledge graph.
• Produces an in-IDE dashboard with two graph views: structural (file/function/class relationships) and domain (business flows and process steps).
• Multi-platform: Claude Code native, Codex, Cursor, VS Code + Copilot, Gemini CLI, OpenCode, Antigravity, and others.
• Incremental updates — only re-analyzes changed files since last run; file analyzers run in parallel (up to 5 concurrent, 20–30 files/batch).
• Guided tours, diff impact analysis, fuzzy+semantic search, persona-adaptive UI, and architectural layer visualisation.
• No token reduction figures claimed; value proposition is developer onboarding and codebase comprehension.
• 8,081 GitHub stars (as of 2026-04-10); MIT license.

What’s novel / different

Unlike tools that focus on minimising tokens sent to the LLM (e.g. context-mode, codebase-memory-mcp, codegraph), Understand-Anything targets the developer’s mental model rather than agent token budgets. The dual structural/domain graph — where the domain view maps code to business processes, flows, and steps — is not offered by any other tool in this category. The persona-adaptive UI (adjusts detail level for junior dev, PM, or power user) and auto-generated guided tours that order architectural concepts by dependency are distinctive product decisions oriented toward human comprehension. The 6th agent (domain-analyzer, triggered by /understand-domain) that extracts business domain ontology from code is the most differentiated capability.

Architecture overview

Core design

Multi-agent pipeline invoked via slash command:

Agent	Role
project-scanner	Discover files, detect languages and frameworks
file-analyzer	Extract functions, classes, imports; produce graph nodes and edges
architecture-analyzer	Identify architectural layers
tour-builder	Generate guided learning tours
graph-reviewer	Validate graph completeness and referential integrity
domain-analyzer	Extract business domains, flows, and process steps (via /understand-domain)

File analyzers run in parallel (up to 5 concurrent, 20–30 files/batch). Output is an interactive dashboard embedded in the IDE.

Interface / API

• Slash commands: /understand, /understand-domain
• MCP server installation for Claude Code and other platforms
• Dashboard: structural graph + domain graph; fuzzy+semantic search; diff impact analysis; guided tours

Dependencies

• Runtime: TypeScript/Node.js
• Multi-platform: MCP protocol for Claude Code, Cursor, Gemini CLI; language server-style integration for VS Code + Copilot
• Storage: local (exact format not documented in README)

Scope / limitations

• No explicit list of supported languages; tool relies on LLM-based file analysis rather than deterministic AST parsing (implies LLM API calls during indexing).
• graph-reviewer by default runs inline (not full LLM review); --review flag needed for thorough validation.
• No token count or performance benchmarks in README.
• Domain graph quality depends on LLM’s ability to infer business semantics from code — unreliable for poorly named or undocumented codebases.

Deployment model

• Runtime: Node.js, local machine
• Install: Claude Code native (claude mcp add), or platform-specific MCP config
• Multi-platform auto-install documented in README

Self-reported metrics

• No quantitative token reduction or speed claims in README.
• Incremental update capability stated (re-analyzes changed files only), no latency figures given.

Open questions

• Whether indexing requires live LLM API calls (implies ongoing cost and latency per /understand run) — not stated.
• Supported language list not specified; claim of “any codebase” is unverified.
• Dashboard persistence format not documented — unclear if graph survives IDE restarts.
• No benchmark against human or automated code comprehension tasks.