saroircommunity/spiderbrain-v3

SpiderBrain v3 is a tool that analyzes your software project and builds a structured, queryable map of it, which AI coding assistants like Claude or Cursor can consult to understand the codebase without having to read through every file from scratch on each request. The README describes the problem it solves as "project Alzheimer's": the gradual loss of context about why files exist, what depends on what, and why past decisions were made. When you run SpiderBrain on a project, it builds a brain folder next to your code. That folder contains a single JSON graph of every file (called a node) and every dependency relationship (called a synapse). Each node gets two scores: a mass score representing how many other files depend on it and how badly things would break if it changed, and a rhythm score tracking how frequently the file is modified. Files with high mass and low rhythm are flagged as masters, meaning they are structural anchors the rest of the project depends on. The brain folder also includes readable markdown summaries ranked by importance and recency, a cluster view grouping related files, a curated changelog for recording why decisions were made, and a deploy log. A cascade engine lets you simulate what would break if a specific file had a fault, tracing the failure along the dependency graph and reporting the blast radius before you make a change. Curated entries like importance overrides, cluster rules, and the changelog survive rebuilds because they are kept separately from the auto-generated graph. The tool integrates with AI tools in two ways. For Claude Code, it wires into session and per-prompt hooks so the brain context is automatically included when the AI starts working. For other tools including Cursor, Zed, and Continue, it runs as an MCP server that exposes four on-demand functions: a brief for per-prompt context, a file lookup, the cascade blast-radius check, and a drift audit. No npm packages are required, only Node 18 or newer. The core engine is licensed under BUSL 1.1, which allows free personal, educational, and open-source use but requires a commercial license for hosted or deployed products. The platform adapters are Apache 2.0. The README includes measured token-reduction benchmarks from real projects, including a 77 percent reduction in tool calls on a 154-node Next.js project.