# CodeGraphContext — Graph Index for Code + MCP

> CodeGraphContext indexes a repo into a code graph so developers and agents can query call chains, dependencies, and architecture via CLI or MCP mode.

## Install

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

## Quick Use

1. Install:
   ```bash
   pip install codegraphcontext
   ```
2. If the CLI isn’t on PATH:
   ```bash
   curl -sSL https://raw.githubusercontent.com/CodeGraphContext/CodeGraphContext/main/scripts/post_install_fix.sh | bash
   ```
3. Initialize and index a repo:
   ```bash
   cd /path/to/repo
   codegraphcontext init
   codegraphcontext index
   ```

## Intro

CodeGraphContext indexes a repo into a code graph so developers and agents can query call chains, dependencies, and architecture via CLI or MCP mode.

- **Best for:** large codebases where agents need reliable call graphs and dependency queries
- **Works with:** Python 3.10+; CLI plus MCP mode for IDE/agent integration
- **Setup time:** 20–45 minutes

## Practical Notes

- Index once, then query call chains/dependencies instead of re-reading files repeatedly.
- Start with a small subfolder first to validate performance before indexing monorepos.

## Two modes: CLI for humans, MCP for agents

You can use CodeGraphContext as:
- a **CLI** to ask structural questions (callers/callees, dependencies),
- an **MCP server** so your assistant can query structure via tools.

## Practical workflow for better agent context

### 1) Index your repo

```bash
codegraphcontext index
```

### 2) Ask “structure” questions first

Good first questions:
- “What calls this function?”
- “Where does this module get instantiated?”
- “Show the call chain from HTTP handler to DB write.”

### 3) Feed the query results to your agent

Instead of dumping full files, give the agent:
- the top 10 nodes in the call chain,
- the 3–5 files it must open,
- constraints (touch only X modules, avoid breaking API).

## Operational advice

- Use file watching mode during refactors only when needed (it updates the graph continuously).
- Treat the graph DB as a cache: rebuild when your repo changes dramatically.

### FAQ

**Q: Do I need MCP to get value?**
A: No. The CLI alone helps answer structural questions quickly.

**Q: When should I re-index?**
A: After large refactors or when results look stale; treat the graph as a cache.

**Q: How does this help agents?**
A: It gives reliable call/dependency context so agents stop guessing where logic lives.

## Source & Thanks

> Source: https://github.com/CodeGraphContext/CodeGraphContext
> License: MIT
> GitHub stars: 3,218 · forks: 573

---

<!-- ZH -->

## 快速使用

1. 安装：
   ```bash
   pip install codegraphcontext
   ```
2. 如果命令找不到（PATH 问题）：
   ```bash
   curl -sSL https://raw.githubusercontent.com/CodeGraphContext/CodeGraphContext/main/scripts/post_install_fix.sh | bash
   ```
3. 初始化并索引仓库：
   ```bash
   cd /path/to/repo
   codegraphcontext init
   codegraphcontext index
   ```

## 简介

CodeGraphContext 会把仓库索引成代码图谱，开发者与 agent 可用 CLI 或 MCP 方式查询调用链、依赖关系与架构视图，从而减少“靠猜”与重复读文件导致的上下文噪音与返工。

- **适合谁：** 代码规模大、agent 需要稳定的调用图/依赖查询能力的工程
- **可搭配：** Python 3.10+；既提供 CLI，也支持 MCP 模式用于 IDE/agent 集成
- **准备时间：** 20–45 分钟

## 实战建议

- 先索引一次，再用调用链/依赖查询替代反复读文件。
- 单体仓库/monorepo 建议先索引子目录验证性能，再全量扩展。

## 两种模式：给人用的 CLI，给 agent 用的 MCP

CodeGraphContext 可以当作：
- **CLI 工具**：做结构化查询（调用者/被调用者、依赖关系等），  
- **MCP Server**：让你的助手通过工具接口查询结构。

## 用于增强 agent 上下文的实战流程

### 1）先索引仓库

```bash
codegraphcontext index
```

### 2）优先问“结构问题”

最有价值的起手问题：
- “谁调用了这个函数？”  
- “这个模块在哪里被实例化？”  
- “从 HTTP handler 到 DB 写入的调用链是什么？”

### 3）把查询结果喂给 agent，而不是丢整库

给 agent 的最好输入通常是：
- 调用链上的关键节点（top 10），  
- 必须打开的 3–5 个文件，  
- 约束条件（只改哪些模块、不要破坏哪些接口）。

## 运维建议

- refactor 期间按需打开 watch 模式（持续更新图谱）。  
- 把图数据库当缓存：当代码结构大幅变化时重建索引即可。

### FAQ

**一定要用 MCP 才有价值吗？**
答：不需要。只用 CLI 做结构查询也能大幅提效。

**什么时候需要重建索引？**
答：大重构后或结果明显过期时；把图谱当缓存即可。

**它怎么帮助 agent？**
答：提供可靠的调用/依赖上下文，减少 agent “猜逻辑在哪”的成本。

## 来源与感谢

> Source: https://github.com/CodeGraphContext/CodeGraphContext
> License: MIT
> GitHub stars: 3,218 · forks: 573


---
Source: https://tokrepo.com/en/workflows/codegraphcontext-graph-index-for-code-mcp
Author: MCP Hub