# code-review-graph — MCP Context for Smarter Reviews

> code-review-graph builds a Tree-sitter code graph and exposes minimal review context via MCP; verified 16,364★ and claims ~8.2× token reduction on 6 repos.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

```bash
pip install code-review-graph  # or: pipx install code-review-graph
code-review-graph install --platform codex  # or: cursor | claude-code | copilot
code-review-graph build
```

## Intro

code-review-graph builds a Tree-sitter code graph and exposes minimal review context via MCP; verified 16,364★ and claims ~8.2× token reduction on 6 repos.

**Best for:** Teams using coding agents who want smaller context windows and clearer blast-radius reviews

**Works with:** Codex, Claude Code, Cursor, Windsurf, Zed, Copilot, and other MCP-capable tools

**Setup time:** 5-12 minutes

### Key facts (verified)

- GitHub: 16380 stars · 1781 forks · pushed 2026-05-08.
- License: MIT · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `code-review-graph`.

## Main

- Run `install` once per machine to generate MCP config, then restart your tool so it can discover the server.
- Use the graph as a review gate: ask for blast radius + minimal file set before edits, then iterate with smaller diffs.
- Use README timings as sanity checks: ~10s initial build for ~500 files; incremental updates under ~2s for a 2,900-file repo (per docs).

### Source-backed notes

- README says it parses repos with Tree-sitter, stores a graph, and serves context through MCP for agent review loops.
- Quick Start shows `pip install` + `install` + `build` as the minimal loop.
- README lists per-platform flags like `--platform codex` / `cursor` / `claude-code` / `copilot-cli`.

### FAQ

- **Do I need MCP support in my tool?**: Yes—your editor/agent must be able to load MCP servers, or you’ll only use the CLI locally.
- **Will it read my entire repo every time?**: The stated goal is the opposite: build once, then incrementally update and serve minimal context.
- **What’s the fastest way to get started?**: Use pipx, run `install --platform <tool>`, restart the tool, then run `build`.

## Source & Thanks

> Source: https://github.com/tirth8205/code-review-graph
> License: MIT
> GitHub: ⭐ 16380 · forks 1781

Thanks to the upstream maintainers for shipping the original project.

---

<!-- ZH -->

## 快速使用

```bash
pip install code-review-graph  # or: pipx install code-review-graph
code-review-graph install --platform codex  # or: cursor | claude-code | copilot
code-review-graph build
```

## 简介

code-review-graph 用 Tree-sitter 构建代码结构图，并通过 MCP 提供最小审阅上下文；已验证 16,364★，并在文档中给出 6 个仓库平均约 8.2× 的 token 节省。

**适合谁用：** 使用 coding agent、想减少上下文与提升“影响面”审阅质量的团队

**适用环境：** Codex、Claude Code、Cursor、Windsurf、Zed、Copilot 等支持 MCP 的工具

**配置耗时：** 5-12 minutes

### 关键事实（已验证）

- GitHub：16380 stars · 1781 forks；最近更新 2026-05-08。
- 许可证：MIT；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中可对照的入口命令：`code-review-graph`。

## 正文

- 每台机器先跑一次 `install` 生成 MCP 配置，然后重启编辑器/Agent CLI，让它能发现并连接 server。
- 把“结构图”当作审阅门禁：先让 agent 计算影响面与最小必读文件集，再开始修改，迭代更省 token。
- 用 README 的时间数据做自检：约 500 文件初次构建 ~10s；约 2,900 文件增量更新 <2s（文档口径）。

### 基于来源的要点

- README 描述其用 Tree-sitter 解析仓库、构建结构图，并通过 MCP 服务为审阅/修改闭环提供最小上下文。
- Quick Start 给出最小闭环：`pip install` → `install` → `build`。
- README 列出多平台安装方式，并提供 `--platform codex/cursor/claude-code/copilot-cli` 等参数。

### 常见问题

- **必须要工具支持 MCP 吗？**：是的。需要编辑器/Agent CLI 能加载 MCP server；否则只能本地 CLI 使用。
- **每次都会全量读取仓库吗？**：文档目标相反：首次构建后增量更新，并在审阅时只返回最小必读上下文。
- **最快的上手路径？**：用 pipx 安装后执行 `install --platform <tool>`，重启工具，再运行 `build`。

## 来源与感谢

> Source: https://github.com/tirth8205/code-review-graph
> License: MIT
> GitHub: ⭐ 16380 · forks 1781


---
Source: https://tokrepo.com/en/workflows/code-review-graph-mcp-context-for-smarter-reviews
Author: MCP Hub