ts-review-graph
Enables Claude Code to query a TypeScript dependency graph and retrieve the minimal set of files needed for code review, implementation, or debugging.
README
ts-review-graph
Build a TypeScript dependency graph in SQLite and tell Claude Code (via MCP) the minimal file set to read before any code review, implementation, or debugging session.
日本語: TypeScript プロジェクトの依存グラフを SQLite に構築し、コードレビュー・実装・デバッグ前に「読むべき最小ファイルセット」を Claude Code (MCP) に伝えるツール。
Why?
Claude Code reads too many files. It grabs large files one after another and stuffs unrelated code into its context.
ts-review-graph pre-builds the project's dependency graph and instantly computes the blast radius (impact surface) of changed files — so Claude only reads what matters.
Real-world benchmark (manako project)
From BENCHMARK.md:
| Metric | Baseline | ts-review-graph | Reduction |
|---|---|---|---|
Read tool calls |
14 | 3 | −79% |
| File content bytes | 219,139 | 58,583 | −73% |
| Estimated tokens | ~54,784 | ~14,645 | −73% |
Graph size: 1,191 nodes / 1,400+ edges (Cloudflare Workers monorepo)
Requirements
- Node.js 20+
Installation
Single tsconfig
npx @elchika-inc/ts-review-graph@latest install --tsconfig tsconfig.json
Config is saved to .ts-review-graph/config.json, MCP server is registered in .mcp.json, and usage instructions are appended to CLAUDE.md. Restart Claude Code and the MCP server connects automatically.
Monorepo (multiple tsconfigs)
npx @elchika-inc/ts-review-graph@latest install \
--tsconfig apps/api/tsconfig.json \
--tsconfig apps/web/tsconfig.app.json \
--tsconfig apps/worker/tsconfig.json
Multiple tsconfigs are merged into a single unified graph (tested with 1,191+ nodes across layers).
Usage
In Claude Code (automatic)
Claude calls get_minimal_context automatically before reading source files:
get_minimal_context(["src/routes/monitors.ts"], "implement")
Example output (implement mode, apps/web + apps/api):
Changed: src/routes/monitors.ts
── Files affected (REVERSE depth=3) ──
1. src/routes/services.ts [IMPORTS_FROM]
── Files to change together (FORWARD depth=1) ──
1. src/env.ts [direct import]
2. src/lib/schemas.ts [direct import]
3. src/lib/format.ts [direct import]
4. packages/db/src/index.ts [direct import]
SKIP: 1170 other files — not in blast radius
CLI commands
| Command | Description |
|---|---|
npx @elchika-inc/ts-review-graph@latest install --tsconfig <path> |
Setup + initial build + register MCP + append CLAUDE.md |
npx @elchika-inc/ts-review-graph build [--tsconfig <path>]... [--db <path>] |
Rebuild the graph |
npx @elchika-inc/ts-review-graph update <file> [--db <path>] |
Incremental update for a single file |
npx @elchika-inc/ts-review-graph status [--db <path>] |
Show graph statistics |
npx @elchika-inc/ts-review-graph uninstall |
Remove MCP registration |
MCP tools
| Tool | Key args | Description |
|---|---|---|
get_minimal_context |
changed_files[], mode (default: "review") |
Minimal file set (REVERSE/FORWARD BFS) |
get_impact |
changed_file |
Files affected by a change, with depth |
get_type_usages |
type_name |
Nodes that reference a type |
get_test_coverage |
file |
Corresponding test files |
query_graph |
from, edge_kind, direction, depth |
General-purpose graph traversal |
build_graph |
tsconfigs[] (optional) |
Rebuild the graph |
graph_status |
— | Graph statistics |
BFS depth by mode
| Mode | REVERSE | FORWARD | Use case |
|---|---|---|---|
review |
depth=2 | — | Pre-review impact analysis (downstream) |
implement |
depth=3 | depth=1 (direct imports only) | Pre-implementation change surface (bidirectional) |
debug |
depth=5 | — | Wide exploration for bug investigation |
Configuration
.ts-review-graph/config.json (auto-generated by install, commit to repo):
{
"tsconfigs": [
"apps/api/tsconfig.json",
"apps/web/tsconfig.app.json",
"apps/monitor-worker/tsconfig.json"
]
}
graph.db is a build artifact — added to .gitignore automatically. Share config.json with your team.
How it works
Build phase
- Load each
tsconfig.json - Walk the AST using the TypeScript Compiler API
- Extract import/export/type relationships
- Store nodes and edges in SQLite (
graph.db)
Query phase
- Specify changed files (
["src/routes/monitors.ts"]) - Run BFS in the selected mode:
- REVERSE: who imports this file (downstream impact)
- FORWARD: what this file imports (upstream co-change candidates)
- Return only files within the blast radius
Tech stack
- Language: TypeScript 5.4+
- Database: SQLite 3 (better-sqlite3)
- Graph traversal: SQL recursive CTE (
WITH RECURSIVE) - MCP: Model Context Protocol SDK v1.0.0
- CLI: Commander.js 12.0.0
License
MIT
Packages
| Package | Description | Version |
|---|---|---|
@elchika-inc/ts-review-graph |
CLI tool | 0.3.0 |
@elchika-inc/ts-review-graph-mcp-server |
MCP server | 0.3.0 |
@elchika-inc/ts-review-graph-core |
Graph build & query engine | 0.3.0 |
Links
- BENCHMARK.md — benchmark data
- CHANGELOG.md — release notes
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。