Kimi Search MCP
# Kimi Search MCP
[English](README.en.md)
非官方的本地 MCP Server,使用开放平台 API Key 调用 Kimi 联网搜索服务。仅支持 stdio。
## 环境要求
- Node.js 22 或更新版本
- npm
- Moonshot 开放平台 API Key
## 安装与启动
推荐使用 npm 全局安装,适合日常使用,启动开销较低且无需自行构建。npx 配置方便,适合快速试用,但启动开销较高;源码构建适合需要修改代码的用户。
以下以 Kimi Code 桌面版为例:将配置合并到用户级 `~/.kimi-code/mcp.json`,替换 API Key 占位符。
保存配置后,如工具未出现在 MCP 列表中,或配置变更未生效,可尝试重启客户端。
### npm 全局安装(推荐)
```shell
npm install -g kimi-search-mcp@latest
```
安装完成后,使用以下启动配置:
- Windows:
```json
{
"mcpServers": {
"kimi-search": {
"command": "cmd",
"args": ["/c", "kimi-search-mcp"],
"env": {
"MOONSHOT_API_KEY": "YOUR_MOONSHOT_API_KEY",
"KIMI_ENABLE_WEB_FETCH": "false"
},
"toolTimeoutMs": 80000
}
}
}
```
- macOS / Linux:
```json
{
"mcpServers": {
"kimi-search": {
"command": "kimi-search-mcp",
"args": [],
"env": {
"MOONSHOT_API_KEY": "YOUR_MOONSHOT_API_KEY",
"KIMI_ENABLE_WEB_FETCH": "false"
},
"toolTimeoutMs": 80000
}
}
}
```
### npx 快速试用
无需全局安装或自行编译,配置方便,但启动开销高于全局安装。
- Windows:
```json
{
"mcpServers": {
"kimi-search": {
"command": "cmd",
"args": ["/c", "npx", "-y", "kimi-search-mcp@latest"],
"env": {
"MOONSHOT_API_KEY": "YOUR_MOONSHOT_API_KEY",
"KIMI_ENABLE_WEB_FETCH": "false"
},
"toolTimeoutMs": 80000
}
}
}
```
- macOS / Linux:
```json
{
"mcpServers": {
"kimi-search": {
"command": "npx",
"args": ["-y", "kimi-search-mcp@latest"],
"env": {
"MOONSHOT_API_KEY": "YOUR_MOONSHOT_API_KEY",
"KIMI_ENABLE_WEB_FETCH": "false"
},
"toolTimeoutMs": 80000
}
}
}
```
首次启动需要联网下载包。
### 从源码构建
从 [GitHub 仓库](https://github.com/CaelixOrbit/kimi-search-mcp) 下载源码,在项目根目录执行:
```shell
npm ci
npm run build
```
构建产物位于 `dist`。将下方配置合并到客户端的 MCP 配置中,替换 API Key 占位符。下面以 Windows 为例,请填写实际绝对路径:
```json
{
"mcpServers": {
"kimi-search": {
"command": "node",
"args": ["C:/path/to/kimi-search-mcp/dist/index.js"],
"env": {
"MOONSHOT_API_KEY": "YOUR_MOONSHOT_API_KEY",
"KIMI_ENABLE_WEB_FETCH": "false"
},
"toolTimeoutMs": 80000
}
}
}
```
## 工具
| MCP 工具 | 对应接口 |
| ------------------- | -------------------------------------------------------------------- |
| `kimi_search_basic` | [联网搜索 Basic](https://platform.kimi.com/docs/api/tools-search) |
| `kimi_search_pro` | [联网搜索 Pro](https://platform.kimi.com/docs/api/tools-search-pro) |
| `kimi_web_fetch` | [网页抓取](https://platform.kimi.com/docs/api/tools-fetch),默认关闭 |
接口区别、参数和使用建议见官方[联网搜索最佳实践](https://platform.kimi.com/docs/guide/best-practices-for-web-search)。
## 环境变量
| 变量 | 规则 |
| ----------------------- | -------------------------------------------- |
| `MOONSHOT_API_KEY` | 必填,使用开放平台 API Key |
| `KIMI_ENABLE_WEB_FETCH` | 控制是否启用 `kimi_web_fetch` 工具,默认关闭 |
考虑到不少 Agent 软件已经自带免费的网页抓取(web_fetch)功能,需要付费使用的 `kimi_web_fetch` 工具默认关闭,也不会出现在客户端的工具列表中。需要时,可将 `KIMI_ENABLE_WEB_FETCH` 设置为 `"true"` 来开启。
## 费用
调用使用开放平台账户额度,具体价格和计费规则以官方[联网搜索与网页抓取定价](https://platform.kimi.com/docs/pricing/websearch)为准。
## 许可证
MIT,Copyright (c) 2026 Caelix。
TDQS
Scored across 2 tools
The two tools are clearly differentiated by tier: basic returns simple results while pro returns text chunks with relevance scores. An agent can distinguish them, though choosing between basic and pro may still require judgment about cost/quality tradeoffs.
Both tools follow the same predictable pattern kimi_search_basic and kimi_search_pro, using a consistent prefix plus a clear tier suffix.
Only two tools make the surface thin, but for a focused search server offering two quality tiers this is borderline rather than inappropriate. It earns each tool's place but leaves little room for parameterization or content retrieval.
For a web search server, basic and pro tiers cover the core capability, including richer output via relevance scores. Minor gaps exist, such as a dedicated content-fetch or pagination tool, but agents can work around them.