**code-review-graph** (CRG) attacks a specific and measurable failure mode of AI coding tools: on review tasks they tend to re-read large portions of a repository, burning tokens on files that have nothing to do with the change. CRG parses the codebase into an AST with **Tree-sitter**, stores it as a graph of nodes (functions, classes, imports) and edges (calls, inheritance, test coverage) in SQLite, and serves precise slices of it to the assistant over **MCP**. The central mechanism is **blast-radius analysis**. When a file changes, the graph traces every caller, dependent, and test that could be affected, and the AI reads only that set instead of scanning the project. The README publishes concrete numbers with a reproduction guide: reading Flask's whole corpus costs 143,594 tokens against 2,196 for a graph-shaped answer — roughly a 71x reduction — with a separate 208,821-token measurement on CRG's own repository. The benchmarks are documented in `docs/REPRODUCING.md` rather than asserted, which is more than most tools in this space offer. Incremental updates are what make it usable day to day. With hooks or watch mode enabled, file saves and commit hooks trigger a diff: CRG finds dependents through the graph's own import and call edges and re-parses only files whose SHA-256 hash actually changed. On a ~3,000-file project (django) a two-file edit re-indexes in about **2.5 seconds**, of which ~1.4s is process start-up — so a no-op update costs essentially just start-up. An initial build runs about 10 seconds for a 500-file project. Distribution is unusually broad for a single-maintainer tool. One `code-review-graph install` command auto-detects which AI coding tools are present and writes the correct MCP configuration for each, covering Codex, Claude Code, Cursor, Windsurf, Zed, Continue, OpenCode, Antigravity, Gemini CLI, Qwen, Qoder, Kiro, CodeBuddy, and GitHub Copilot including its CLI. It also installs platform-native hooks and skills where supported and injects graph-aware instructions into platform rules. The uninstall path gets equal attention, which is a good signal for a tool that edits shared config: `uninstall` is symmetric, supports `--dry-run`, removes only CRG-owned files and entries while leaving unrelated MCP servers, hooks, and JSONC comments untouched, and uses atomic replacement so a failed write leaves the original intact. Requires Python 3.10+, ships on PyPI, MIT-licensed, with **v2.3.7** the current tag and README translations in five languages.