# codedb — MCP-Native Code Intelligence Server

> Zig-powered code intelligence server with 16 MCP tools + HTTP/CLI for indexing, search, and dep graphs; verified 797★, pushed 2026-05-12.

## Install

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

## Quick Use

```bash
curl -fsSL https://codedb.codegraff.com/install.sh | bash
codedb mcp /path/to/your/project
codedb serve /path/to/your/project  # localhost HTTP
```

## Intro

Zig-powered code intelligence server with 16 MCP tools + HTTP/CLI for indexing, search, and dep graphs; verified 797★, pushed 2026-05-12.

**Best for:** Agents that need fast repo-wide search, outline, and dependency context without scanning files every turn

**Works with:** macOS/Linux; installs via `install.sh` and registers as an MCP server for Claude Code/Codex/Gemini/Cursor (per README)

**Setup time:** 10-22 minutes

### Key facts (verified)

- GitHub: 797 stars · 52 forks · pushed 2026-05-12.
- License: BSD-3-Clause · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `curl -fsSL https://codedb.codegraff.com/install.sh | bash`.

## Main

- Index once, query many: use `codedb mcp <repo>` so agents can ask for outlines/search/dep graphs without repeated full scans.

- Keep privacy boundaries: README mentions sensitive file blocking (e.g., `.env`); still review what you index on shared machines.

- Use HTTP mode for debugging and MCP stdio for production agent use—keep the tool surface consistent.

- Prefer structural tools over raw grep: outlines and dependency graphs often answer “where is this used?” faster than text search.

### Source-backed notes

- README describes a Zig core with MCP-native support and lists 16 MCP tools for codebase intelligence.
- README install is a one-liner `install.sh` that auto-registers the MCP server for multiple AI CLIs.
- README shows quick start for MCP (`codedb mcp`) and HTTP server (`codedb serve`).

### FAQ

- **Do I need to build from source?**: No — README provides a one-line installer that downloads release binaries and configures MCP.
- **Is it safe for secrets?**: README mentions sensitive file blocking, but you should still avoid indexing secret-heavy folders.
- **How do agents use it?**: Run `codedb mcp <repo>` and then call MCP tools for search/outline/graph from your MCP client.

## Source & Thanks

> Source: https://github.com/justrach/codedb
> License: BSD-3-Clause
> GitHub stars: 797 · forks: 52

---

<!-- ZH -->

## Quick Use

```bash
curl -fsSL https://codedb.codegraff.com/install.sh | bash
codedb mcp /path/to/your/project
codedb serve /path/to/your/project  # localhost HTTP
```

## Intro

codedb 是 Zig 编写的代码智能服务器：16 个 MCP 工具 + 本地 HTTP/CLI，用于索引、搜索与依赖图，并支持自动注册到常见 AI CLI；已验证 797★，更新于 2026-05-12。

**Best for:** 希望 agent 以更低 token 成本做全仓搜索/大纲/依赖分析的团队

**Works with:** macOS/Linux；README 提供 `install.sh` 并可自动注册到 Claude Code/Codex/Gemini/Cursor 的 MCP 配置中

**Setup time:** 10-22 minutes

### Key facts (verified)

- GitHub：797 stars · 52 forks；最近更新 2026-05-12。
- 许可证：BSD-3-Clause；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：`curl -fsSL https://codedb.codegraff.com/install.sh | bash`。

## Main

- 一次索引，多次查询：用 `codedb mcp <repo>` 让 agent 直接拿到 outline/search/依赖图，避免每轮都扫全仓。

- 注意隐私边界：README 提到会屏蔽敏感文件（如 `.env`），但共享机器上仍建议明确索引范围与目录白名单。

- 调试用 HTTP，生产用 MCP stdio：把交互面稳定下来，便于接入不同 agent 工具链。

- 优先用结构化能力而不是纯 grep：outline/依赖图往往更快回答“哪里引用了它”。

### Source-backed notes

- README 说明其 Zig 核心并原生支持 MCP，列出 16 个代码智能相关的 MCP 工具。
- README 提供一行 `install.sh` 安装，并可自动把 MCP server 注册到多个 AI CLI。
- README 给出 MCP（`codedb mcp`）与 HTTP（`codedb serve`）两种启动方式。

### FAQ

- **需要源码编译吗？**：不需要；README 给出一行安装脚本，会下载 release 二进制并配置 MCP。
- **对密钥安全吗？**：README 提到会屏蔽敏感文件，但仍建议避免索引包含大量密钥的目录。
- **agent 怎么用？**：启动 `codedb mcp <repo>`，然后在 MCP 客户端里调用搜索/大纲/依赖图等工具。

## Source & Thanks

> Source: https://github.com/justrach/codedb
> License: BSD-3-Clause
> GitHub stars: 797 · forks: 52


---
Source: https://tokrepo.com/en/workflows/codedb-mcp-native-code-intelligence-server
Author: MCP Hub