# Claude Context — Semantic Code Search MCP

> Claude Context is an MCP server for semantic code search, so agents can retrieve relevant code without stuffing whole folders into the prompt.

## Install

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

## Quick Use

1. Install / set up:
   ```bash
   claude mcp add claude-context \
+  -e OPENAI_API_KEY=sk-your-openai-api-key \
+  -e MILVUS_ADDRESS=your-zilliz-cloud-public-endpoint \
+  -e MILVUS_TOKEN=your-zilliz-cloud-api-key \
+  -- npx @zilliz/claude-context-mcp@latest
   ```
2. Start / smoke test:
   ```bash
   # In Claude Code
/mcp
   ```
3. Verify:
   - Index one repo and run a natural-language search (e.g., auth handlers). Confirm results are returned and referenced in answers.

## Intro

Claude Context is an MCP server for semantic code search, so agents can retrieve relevant code without stuffing whole folders into the prompt.

- **Best for:** Teams with medium/large codebases who want retrieval-first context for agents
- **Works with:** Claude Code MCP; runs via `npx @zilliz/claude-context-mcp@latest` with env vars
- **Setup time:** 15 minutes

## Practical Notes

- Setup time ~15 minutes (configure env + add MCP server + index once)
- Measurable check: 1 indexed codebase + successful `search_code` results
- GitHub stars + forks (verified): see Source & Thanks

This is a good fit when you find yourself doing expensive “context stuffing”:

- Instead of pasting directories, index once and retrieve by query.
- Prefer a repeatable workflow: index → search → cite code locations.
- Keep the search layer separate from generation so you can swap models without re-indexing.

Operationally, start with one repo and one embedding provider, then expand once the team trusts retrieval quality.

### FAQ

**Q: Do I need Node.js?**
A: Yes — the README lists Node.js 20+ for the MCP server run via npx.

**Q: What credentials are required?**
A: You configure env vars for embeddings and your vector DB endpoint/token as documented.

**Q: Where do I use it?**
A: Add it as an MCP server to Claude Code (or another MCP-capable client).

## Source & Thanks

> Source: https://github.com/zilliztech/claude-context
> License: MIT
> GitHub stars: 10,953 · forks: 822

---

<!-- ZH -->

## 快速使用

1. 安装 / 设置：
   ```bash
   claude mcp add claude-context \
+  -e OPENAI_API_KEY=sk-your-openai-api-key \
+  -e MILVUS_ADDRESS=your-zilliz-cloud-public-endpoint \
+  -e MILVUS_TOKEN=your-zilliz-cloud-api-key \
+  -- npx @zilliz/claude-context-mcp@latest
   ```
2. 启动 / 冒烟测试：
   ```bash
   # In Claude Code
/mcp
   ```
3. 验证：
   - 对一个仓库完成索引，并用自然语言搜索（例如认证相关函数）；确认能返回结果并被引用到回答里。

## 简介

Claude Context 是一套 MCP 代码搜索服务：为 Claude Code 等 agent 提供语义检索，把相关代码按需取回进上下文，减少整目录喂给模型的成本，并提升大仓库命中率。

- **适合谁：** 中大型代码库团队，希望让 agent 先检索再回答，而不是盲读全仓
- **可搭配：** Claude Code 的 MCP；通过 `npx @zilliz/claude-context-mcp@latest` + 环境变量运行
- **准备时间：** 15 分钟

## 实战建议

- 准备时间约 15 分钟（配 env + 添加 MCP + 首次索引）
- 可量化检查：至少 1 个仓库索引完成，并能成功返回 `search_code` 结果
- GitHub stars / forks（已核验）：见「来源与感谢」

当你经常被“塞上下文”拖慢/拖贵时，这类 MCP 检索层很合适：

- 不再粘贴目录：先索引一次，再按查询取回。
- 固化流程：index → search → 引用代码位置。
- 把检索层与生成层解耦：换模型不必重建索引。

落地建议：先从 1 个仓库 + 1 种 embedding provider 开始，等团队认可检索质量再扩展。

### FAQ

**需要 Node.js 吗？**
答：需要。README 写明通过 npx 运行并要求 Node.js 20+。

**需要哪些凭据？**
答：按文档配置 embeddings 与向量库的 env（地址/令牌等）。

**在哪用？**
答：把它添加为 Claude Code（或其他 MCP 客户端）的 MCP server。

## 来源与感谢

> Source: https://github.com/zilliztech/claude-context
> License: MIT
> GitHub stars: 10,953 · forks: 822


---
Source: https://tokrepo.com/en/workflows/claude-context-semantic-code-search-mcp
Author: MCP Hub