# Graphify — Repo Knowledge Graph + MCP

> Graphify extracts docs/code into a knowledge graph and can install as an MCP/skill across Claude Code, Cursor, Codex, and Gemini CLI. Install via uv/pipx.

## Install

Copy the content below into your project:

# Graphify — Repo Knowledge Graph + MCP

> Graphify extracts docs/code into a knowledge graph and can install as an MCP/skill across Claude Code, Cursor, Codex, and Gemini CLI. Install via uv/pipx.

## Quick Use

1) Install

```bash
uv tool install graphifyy && graphify install
```

2) Run

```bash
graphify extract ./docs
```

3) Verify

```bash
graphify --help
```

---

## Intro

**Best for**: teams who need a structured repo map (entities + relationships) and want the agent to cite grounded nodes instead of guessing

**Works with**: Python, uv/pipx installs, optional MCP integration, multiple backends per README

**Setup time**: 11 minutes

### Quant Data

- Supports multiple agent platforms and MCP installs (repo)
- Setup time ~11 minutes

---

## How to Use It Well

Use Graphify as a grounding layer: generate the graph, then ask agents to cite graph nodes when making architectural claims or refactor plans.

### Adoption Checklist

- Start with one real task and keep the scope narrow
- Capture a baseline: time-to-first-success and output quality
- Version your config/skills so teammates stay in sync

### Guardrails

Treat extracted graphs as build artifacts. If the repo changes fast, prefer smaller scope extractions (one module at a time) to control cost.

### FAQ

**Q: Is this only for docs?**
A: No. The README describes extracting from docs/code and optional media formats via extras; start with docs for a fast baseline.

**Q: How do I choose an LLM backend?**
A: Start with the backend you already pay for, then compare extraction quality and latency before switching.

**Q: How do I keep graphs fresh?**
A: Regenerate on meaningful diffs (weekly or per-release) and keep outputs in a dedicated directory so updates are reviewable.

---

## Source & Thanks

> GitHub: https://github.com/safishamsi/graphify
> Owner avatar: https://avatars.githubusercontent.com/u/216348667?v=4
> License (SPDX): MIT
> GitHub stars (verified via `api.github.com/repos/safishamsi/graphify`): 46,535


---

<!-- ZH -->

# Graphify——代码库知识图谱与 MCP

> Graphify 可把代码/文档抽取成知识图谱，并可作为 MCP/skill 接入 Claude Code、Cursor、Codex、Gemini CLI。用 uv 或 pipx 安装后运行 graphify install。

## 快速使用

1）安装

```bash
uv tool install graphifyy && graphify install
```

2）运行

```bash
graphify extract ./docs
```

3）验证

```bash
graphify --help
```

---

## 简介

**适合谁**：需要结构化的代码库地图（实体+关系），希望 agent 基于可引用节点回答而不是猜的团队

**适用环境**：Python、uv/pipx 安装方式、可选 MCP 集成、README 提到多种后端可选

**安装耗时**：11 分钟

### 量化信息

- 支持多种 agent 平台并可安装为 MCP（仓库）
- 装机约 11 分钟

---

## 用好它的方式

把 Graphify 当作 grounding 层：先生成图谱，再要求 agent 在做架构判断/重构计划时引用图谱节点。

### 推广清单

- 先选一个真实任务，小范围试跑
- 记录基线：首次成功耗时与输出质量
- 配置/技能要版本化，避免团队漂移

### 风险与护栏

把抽取结果当成 build 产物：仓库变化快时，优先小范围抽取（按模块），控制成本与噪音。

### FAQ

**Q: 只能处理文档吗？**
A: 不是。README 描述可从文档/代码抽取，并可通过 extras 支持更多格式；建议先从 docs 建立基线。

**Q: LLM 后端怎么选？**
A: 先用你已有订阅/额度的后端，再比较抽取质量与延迟后决定是否切换。

**Q: 怎么保持图谱更新？**
A: 在关键 diff 后重建（按周或发版），把输出放在固定目录，更新就能像代码一样 review。

---

## 来源与感谢

> GitHub：https://github.com/safishamsi/graphify
> Owner avatar：https://avatars.githubusercontent.com/u/216348667?v=4
> 许可证（SPDX）：MIT
> GitHub stars（已通过 `api.github.com/repos/safishamsi/graphify` 核验）：46,535


---
Source: https://tokrepo.com/en/workflows/graphify-repo-knowledge-graph-mcp
Author: Script Depot