MedXpert Global Regulatory Connector (reg-connector)

Read-only local MCP server for global medical-device regulation — 27 knowledge hubs (NMPA / FDA / EU MDR / PMDA + orthopedic global registration paths) exposed as 7 tools, with a built-in freshness guard.

中英双语说明：本文英文在前、中文在后。mcp-name: io.github.zhaoxinghua09-cell/medxpert-reg-connector

English

What it is

A local, read-only MCP (Model Context Protocol) server over a curated medical-device regulatory knowledge base. Any MCP host — Claude Desktop, Cursor, LobeHub, WorkBuddy — can call its tools directly to look up registration paths, classification clues, UDI cross-jurisdiction mappings, and knowledge freshness.

• Zero egress, zero credentials, zero side effects at runtime
• 7 tools, every one annotated readOnlyHint: true
• 27 hubs shipped inside the wheel — pip install is enough, nothing else to download
• Freshness guard: every search result carries as_of / age_days / freshness
• Copyright-safe: paid standards (ISO/IEC/ASTM/GB) are never stored — only numbers and official lookup entry points

Install

# recommended — the knowledge base is bundled, no extra steps
uvx medxpert-reg-connector
# or
pip install medxpert-reg-connector
medxpert-reg-connector

Host config (mcp.json / claude_desktop_config.json):

{
  "mcpServers": {
    "medxpert-reg-connector": {
      "command": "python",
      "args": ["-m", "medxpert_reg_connector"]
    }
  }
}

Docker (no Python setup on the host):

docker build -t medxpert-reg-connector .
docker run -i --rm medxpert-reg-connector

Optional: point REG_HUB_REFS at your own knowledge-base directory to override the bundled one.

Tools

Verified output (real run, 2026-11-15)

search_regulation("UDI", top_k=3)
-> 3 hits, top1 = UDI全球标识枢纽 (CN/US/EU/JP cross-reference: GS1-MA, GUDID, EUDAMED, KikiDB)

kb_freshness()
-> {hubs: 27, ok: 27, aging: 0, stale: 0, unknown: 0, overdue: 27}

Scope and limits

1. Results are knowledge-base leads, not regulatory conclusions — always confirm against the target market's official classification database (NMPA / FDA / MDR).
2. Local and read-only; no network egress at runtime (the polling pipeline refresh_scan.py is a separate, opt-in script).
3. Paid standards are never ingested — only numbers + official lookup entry points.
4. Reload the server after refreshing the knowledge base.

License

• Code: Apache-2.0 (see LICENSE), copyright 赵兴华 / Steven Zhao·China (ORCID 0009-0001-0512-1237).
• Knowledge base (medxpert_reg_connector/data/references/, methodology, taxonomy): All Rights Reserved (see CONTENT-LICENSE) — citation with attribution is allowed; copying, redistribution, resale, or training use is not.
• MedXpert / SynomosAI / LGD are project marks; no legal entity or trademark registration is claimed.
• Not regulatory, legal, or registration-agent advice.

中文

这是什么

把经过整理的医疗器械法规知识库封装成本地只读 MCP 服务器。任意 MCP 客户端（Claude Desktop / Cursor / LobeHub / WorkBuddy）都可直接调用其工具，查询注册路径、分类线索、UDI 多辖区对照与知识库时效。

• 运行时零外发、零凭据、零副作用
• 7 个工具，全部声明 readOnlyHint: true
• 27 个枢纽随 wheel 分发——pip install 装完即可检索，无需另外克隆
• 时效护栏：每条检索结果都带 as_of / age_days / freshness
• 版权安全：付费标准（ISO/IEC/ASTM/GB）永不入库，只存编号与官方查询入口

安装

uvx medxpert-reg-connector        # 推荐：知识库已随包分发
pip install medxpert-reg-connector
medxpert-reg-connector

客户端配置见上方 mcp.json 片段；也可用仓库内 Dockerfile 构建镜像，免装 Python 环境。需要接自建知识库时，用环境变量 REG_HUB_REFS 指向自定义目录。

工具

7 个工具的分工见上方英文表（入参、注解、用途一致）。核心三件套：

• list_hubs / get_hub —— 知识库全景与单枢纽全文
• search_regulation / ask_classification —— 关键词检索与产品分类速查
• kb_freshness / reg_delta / kb_sources —— 时效自检、变更流水、官方源注册表

实跑返回（2026-11-15 实测）

search_regulation("UDI", top_k=3)
-> 命中 3 条，top1 = UDI全球标识枢纽（中/美/欧/日四体系：GS1-MA、GUDID、EUDAMED、KikiDB）

kb_freshness()
-> {hubs: 27, ok: 27, aging: 0, stale: 0, unknown: 0, overdue: 27}

边界

1. 返回为知识库线索，不是法规结论——分类与注册路径须以目标市场官方分类库（NMPA 分类目录 / FDA 产品分类库 / MDR 分类规则）为准。
2. 本地只读，运行时不联网（巡检脚本 refresh_scan.py 为独立、可选的手动流程）。
3. 付费标准全文不入库，只存编号 + 官方查询入口。
4. 知识库更新后需重载 server。

许可与归属

• 代码 Apache-2.0（见 LICENSE），版权归「赵兴华 / Steven Zhao·China（ORCID 0009-0001-0512-1237）」。
• 知识库/方法论/分类体系 保留所有权利（详见 CONTENT-LICENSE）：可注明出处引用，不得复制、再分发、转售或用于模型训练。
• MedXpert / SynomosAI / LGD 为项目标识，未申请实体注册与商标注册；出现仅作来源标识，不构成对法人实体或商标权的任何主张。
• 本仓库内容不构成法规意见、法律意见或注册代理服务；关键数据以监管机构最新发布为准。

联系：zhaoxinghua09@gmail.com ｜ ORCID 0009-0001-0512-1237