chinese-law-mcp
# chinese-law-mcp
中国法律法规 MCP 服务器:让 Claude、Cursor 等 AI 助手直接逐条查询 **1500+ 部现行中国法律、法律解释、司法解释与行政法规**(56,000+ 条)。实时数量以上游语料为准,见 [chinese-law-corpus](https://gitee.com/chengzi_h/chinese-law-corpus)(其 README 计数由脚本生成)。
MCP server for Chinese statutes: search 1500+ Chinese statutes, judicial interpretations and administrative regulations in force and fetch any article verbatim, with citable source links. Install with `npx -y chinese-law-mcp` — no API key, no telemetry, read-only.
```json
{ "mcpServers": { "chinese-law": { "command": "npx", "args": ["-y", "chinese-law-mcp"] } } }
```
## 工具
| Tool | 说明 |
| --- | --- |
| `search_laws` | 按名称搜索法律(如「民法典」「劳动合同」) |
| `get_article` | 取指定条文原文(如民法典第 577 条),带章节位置、施行日期与可引用链接 |
| `search_in_law` | 在某部法律全文内按关键词检索条文 |
## 安装
无需安装,`npx` 直接运行。Claude Desktop / Cursor / Cline 等客户端配置:
```json
{
"mcpServers": {
"chinese-law": {
"command": "npx",
"args": ["-y", "chinese-law-mcp"]
}
}
}
```
容器运行(stdio,不开端口):
```bash
docker build -t chinese-law-mcp .
docker run -i --rm chinese-law-mcp
```
## 配置教程
面向律师的图文配置步骤(含客户端选择、安全注意项与使用示例):
[让 AI 助手直接查到现行法条:给律师的 MCP 入门](https://lttxzmj.github.io/lawpulse-site/blog/mcp-connect-ai-to-law.html)。
更多律师 AI 实务教程见 [律师 AI 实务手册](https://lttxzmj.github.io/lawpulse-site/blog/index.html)。
## 数据与合规
- 数据来自 [chinese-law-corpus](https://gitee.com/chengzi_h/chinese-law-corpus)(Gitee,CC0),经其对象存储直读通道只读拉取,进程内缓存;本服务器无遥测、无写操作。
- 法律文本依《著作权法》第五条不受著作权保护;本项目为非官方整理,正式引用请以官方公布文本为准,工具输出自带此提示。
- 每条结果附网页版链接([法脉法律库](https://lttxzmj.github.io/lawpulse-site/),57,000+ 页逐条可引用)。
## 关于
由 [法脉 LawPulse](https://lttxzmj.github.io/lawpulse-site/) 维护——给中国执业律师的本地优先桌面研究工具(离线法律库 + 法条笔记 + 引用可核实的合同审查)。数据随法律季度更新。
MIT License.
## 相关项目
- [chinese-law-skills](https://github.com/lttxzmj/chinese-law-skills):同一语料的 Agent 技能库(法条引用核验、合同审查清单、诉讼计算器等 13 个,Claude Code 一条命令安装,plugin 已捆带本 MCP)。
- [法脉桌面端](https://lttxzmj.github.io/lawpulse-site/):离线法律库 + 法条笔记 + AI 合同审查。
This server cannot be deployed
TDQS
Scored across 3 tools
search_laws (find a law by name) and search_in_law (keyword search within one law's articles) are superficially similar but the descriptions clearly distinguish scope: one operates over the catalog, the other within a single document. get_article is clearly distinct as a targeted retrieval. Minor risk of misselection only when an agent doesn't read carefully.
All three use snake_case verb_noun-ish patterns: search_laws, get_article, search_in_law. The phrasing between search_laws and search_in_law is slightly asymmetric (plural target vs prepositional scope), but the convention is otherwise predictable and readable.
Three tools is on the thin side for a legal research server covering 412 laws; there's little room for browsing, listing, or metadata exploration. It is not extreme, but it feels minimal rather than well-scoped.
Core retrieval (find law, get article, keyword-in-law) is covered, but notable gaps exist: no way to list/enumerate laws, browse by chapter or category, filter by effective date/domain, or follow cross-references. Agents needing discovery beyond a name search will hit a dead end.