![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120304-688a09f859b06.webp\")
<\/p>\n\u4f5c\u8005：\u6765\u81ea Elastic\u00a0Alex Salgado<\/p>\n
<\/p>\n
\u5b66\u4e60\u5982\u4f55\u4f7f\u7528 FastMCP \u548c Elasticsearch \u6784\u5efa\u4e00\u4e2a MCP \u670d\u52a1\u5668\u6765\u7ba1\u7406\u548c\u641c\u7d22\u6570\u636e\u3002<\/p>\n<\/p>\n
Elasticsearch \u5177\u5907\u4e0e\u4e1a\u754c\u9886\u5148\u7684\u751f\u6210\u5f0f AI \u5de5\u5177\u548c\u670d\u52a1\u5546\u7684\u539f\u751f\u96c6\u6210\u3002\u67e5\u770b\u6211\u4eec\u7684\u7f51\u7edc\u7814\u8ba8\u4f1a，\u4e86\u89e3\u5982\u4f55\u8d85\u8d8a RAG \u57fa\u7840，\u6216\u4f7f\u7528 Elastic \u5411\u91cf\u6570\u636e\u5e93\u6784\u5efa\u53ef\u6295\u5165\u751f\u4ea7\u7684\u5e94\u7528\u3002<\/p>\n
\u4e3a\u4e86\u6784\u5efa\u6700\u9002\u5408\u4f60\u7528\u4f8b\u7684\u641c\u7d22\u89e3\u51b3\u65b9\u6848，\u73b0\u5728\u5c31\u5f00\u59cb\u514d\u8d39\u7684\u4e91\u8bd5\u7528，\u6216\u5728\u672c\u5730\u673a\u5668\u4e0a\u8bd5\u7528 Elastic\u3002<\/p>\n
\u5927\u578b\u8bed\u8a00\u6a21\u578b（LLMs）\u662f\u9769\u547d\u6027\u7684\u5de5\u5177，\u4f46\u5f53\u5b83\u4eec\u62e5\u6709\u5b9e\u65f6\u4e0a\u4e0b\u6587\u5e76\u80fd\u4e0e\u5916\u90e8\u4e16\u754c\u4ea4\u4e92\u65f6，\u5176\u771f\u6b63\u6f5c\u529b\u624d\u80fd\u88ab\u91ca\u653e\u3002\u6a21\u578b\u4e0a\u4e0b\u6587\u534f\u8bae（Model Context Protocol –\u00a0MCP）\u5c31\u50cf\u4e00\u5ea7\u6865\u6881，\u5c06\u4f60\u7684 LLM \u8fde\u63a5\u5230\u771f\u5b9e\u6570\u636e\u548c\u5916\u90e8\u5de5\u5177\u3002\u901a\u8fc7\u63d0\u4f9b\u8fd9\u4e9b\u4e0a\u4e0b\u6587，LLM \u80fd\u591f\u7ed9\u51fa\u66f4\u76f8\u5173\u3001\u66f4\u771f\u5b9e\u7684\u54cd\u5e94\u3002<\/p>\n
\u5c3d\u7ba1\u4f20\u7edf\u7684 RAG（\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u4e3a LLM \u63d0\u4f9b\u4e0a\u4e0b\u6587，\u4f46\u5b83\u901a\u5e38\u4f9d\u8d56\u4e8e\u9884\u5148\u5b58\u50a8\u5728\u6570\u636e\u5e93\u6216\u79c1\u6709\u6570\u636e\u96c6\u4e2d\u7684\u9759\u6001\u4fe1\u606f\u3002\u76f8\u6bd4\u4e4b\u4e0b，MCP \u53ef\u4ee5\u5c06\u6765\u81ea\u6570\u636e\u5e93\u3001API\u3001\u4f20\u611f\u5668\u548c\u5176\u4ed6\u6570\u636e\u6d41\u7b49\u5404\u79cd\u7cfb\u7edf\u7684\u52a8\u6001\u5b9e\u65f6\u4fe1\u606f\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u5f15\u5165\u3002<\/p>\n
\u5728\u8fd9\u7bc7\u535a\u5ba2\u4e2d，\u4f60\u5c06\u5b66\u4e60\u5982\u4f55\u4f7f\u7528 FastMCP \u548c Elasticsearch \u6784\u5efa\u4e00\u4e2a MCP \u670d\u52a1\u5668\u6765\u7ba1\u7406\u548c\u641c\u7d22\u6570\u636e\u3002\u4f5c\u4e3a\u793a\u4f8b，\u6211\u4eec\u5c06\u4f7f\u7528 Apple Watch \u6570\u636e，\u6f14\u793a\u5982\u4f55\u5c06\u539f\u59cb\u4fe1\u606f\u8f6c\u5316\u4e3a\u667a\u80fd\u6d1e\u5bdf\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120304-688a09f8736b0.webp\")
<\/p>\n
\u4f7f\u7528\u81ea\u7136\u8bed\u8a00\u4ee5\u82f1\u6587\u3001\u8461\u8404\u7259\u6587\u6216\u4e2d\u6587\u63d0\u51fa\u5173\u4e8e\u4f60\u6570\u636e\u7684\u590d\u6742\u95ee\u9898\u3002<\/p>\n
MCP \u662f\u4e00\u4e2a\u5f00\u653e\u534f\u8bae，\u7528\u4e8e\u6807\u51c6\u5316 LLMs \u5982\u4f55\u4e0e\u5916\u90e8\u6570\u636e\u6e90\u548c\u5de5\u5177\u4ea4\u4e92\u3002\u60f3\u4e86\u89e3\u8be6\u7ec6\u4ecb\u7ecd，\u8bf7\u53c2\u9605 JD Armada \u7684\u6587\u7ae0： MCP（Model Context Protocol，\u6a21\u578b\u4e0a\u4e0b\u6587\u534f\u8bae）\u7684\u5f53\u524d\u72b6\u6001\u3002<\/p>\n<\/p>\n
\u4f7f\u7528 MCP \u4e3a\u57fa\u4e8e LLM \u7684 Agent \u5f00\u53d1\u5e26\u6765\u591a\u4e2a\u4f18\u52bf：<\/p>\n
\u804c\u8d23\u5206\u79bb：MCP \u5c06\u4e0a\u4e0b\u6587\u903b\u8f91（\u6570\u636e\u548c\u5de5\u5177）\u4e0e LLM \u4ea4\u4e92\u5206\u79bb，\u4f7f\u5f00\u53d1\u66f4\u6709\u6761\u7406\u3002<\/p>\n<\/li>\n
\u6807\u51c6\u5316：\u9075\u5faa\u6807\u51c6\u534f\u8bae，\u5f00\u53d1\u8005\u53ef\u4ee5\u6784\u5efa\u517c\u5bb9\u4efb\u4f55 MCP \u5ba2\u6237\u7aef\u7684\u5de5\u5177，\u65e0\u8bba\u4f7f\u7528\u54ea\u4e2a LLM\u3002<\/p>\n<\/li>\n
\u53ef\u91cd\u7528\u6027：\u5df2\u521b\u5efa\u7684\u5de5\u5177\u548c\u8d44\u6e90\u53ef\u4ee5\u8f7b\u677e\u590d\u7528\u4e8e\u4e0d\u540c\u9879\u76ee\u548c\u4e0a\u4e0b\u6587\u4e2d\u3002<\/p>\n<\/li>\n
\u53ef\u6269\u5c55\u6027：\u901a\u8fc7 MCP，\u53ef\u4ee5\u6a2a\u5411\u6269\u5c55，\u6dfb\u52a0\u66f4\u591a\u670d\u52a1\u5668\u4ee5\u5904\u7406\u4e0d\u540c\u7684\u77e5\u8bc6\u9886\u57df\u6216\u529f\u80fd\u3002<\/p>\n<\/li>\n<\/ul>\n
MCP \u5b9a\u4e49\u4e86\u4e09\u4e2a\u4e3b\u8981\u7ec4\u4ef6，\u7528\u4e8e\u5efa\u7acb LLM \u4e0e\u5916\u90e8\u7cfb\u7edf\u4ea4\u4e92\u7684\u6807\u51c6：<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120304-688a09f8a223b.webp\")
<\/p>\n
1) Resources：LLM \u7684 \u201c\u773c\u775b\u201d<\/p>\n
\u529f\u80fd：\u901a\u8fc7\u76f4\u63a5\u66b4\u9732\u6570\u636e\u4e3a LLM \u63d0\u4f9b\u4e0a\u4e0b\u6587<\/p>\n<\/li>\n
\u884c\u4e3a：\u7c7b\u4f3c REST API \u4e2d\u7684 GET \u7aef\u70b9，\u53ea\u8bfb\u53d6\u6570\u636e\u4e0d\u505a\u4fee\u6539<\/p>\n<\/li>\n
\u793a\u4f8b： health:\/\/steps\/latest – \u83b7\u53d6\u6700\u65b0\u7684\u6b65\u6570\u8bb0\u5f55<\/p>\n<\/li>\n
URI \u6a21\u5f0f：\u4f7f\u7528\u57df\u540d\u547d\u540d\u7a7a\u95f4\u6a21\u5f0f<\/p>\n<\/li>\n<\/ul>\n
2) Tools：LLM \u7684 \u201c\u624b\u201d<\/p>\n
\u529f\u80fd：\u652f\u6301\u6267\u884c\u64cd\u4f5c\u548c\u8ba1\u7b97，\u662f\u5904\u7406\u590d\u6742\u67e5\u8be2\u548c\u5206\u6790\u7684\u5173\u952e<\/p>\n<\/li>\n
\u884c\u4e3a：\u7c7b\u4f3c REST API \u4e2d\u7684 POST \u7aef\u70b9，\u4f1a\u6267\u884c\u64cd\u4f5c，\u53ef\u80fd\u6709\u526f\u4f5c\u7528<\/p>\n<\/li>\n
\u7279\u70b9：<\/p>\n
\u63a5\u53d7\u7ed3\u6784\u5316\u53c2\u6570<\/p>\n<\/li>\n
\u6839\u636e\u53c2\u6570\u6267\u884c\u52a8\u6001\u903b\u8f91<\/p>\n<\/li>\n
\u4e3b\u52a8\u5904\u7406\u548c\u64cd\u4f5c\u6570\u636e，\u800c\u4e0d\u4ec5\u4ec5\u662f\u83b7\u53d6\u6570\u636e<\/p>\n<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n
3) Prompts：LLM \u7684 \u201c\u5730\u56fe\u201d（MCP \u7279\u6709）<\/p>\n
\u529f\u80fd：\u4e3a\u5e38\u89c1\u4efb\u52a1\u63d0\u4f9b\u53ef\u91cd\u7528\u7684\u547d\u4ee4\u6a21\u677f，\u4e0e\u4f20\u7edf\u5c06\u7528\u6237\u8f93\u5165\u548c\u4efb\u52a1\u8bf4\u660e\u6df7\u5408\u7684 Prompt \u4e0d\u540c<\/p>\n<\/li>\n
\u884c\u4e3a：\u4f5c\u4e3a\u5df2\u6ce8\u518c\u6a21\u677f\u51fa\u73b0，\u53ef\u4f5c\u4e3a\u659c\u6760\u547d\u4ee4\u4f7f\u7528<\/p>\n<\/li>\n
\u793a\u4f8b：\/daily_report – \u89e6\u53d1\u9884\u8bbe\u7684\u5206\u6790\u5de5\u4f5c\u6d41<\/p>\n<\/li>\n
\u76ee\u7684：\u89c4\u8303\u5e38\u89c1\u4ea4\u4e92\u4e0e\u5de5\u4f5c\u6d41\u7a0b<\/p>\n<\/li>\n<\/ul>\n
\u5728\u5f00\u59cb\u7f16\u7801\u4e4b\u524d，\u5148\u4e86\u89e3\u4e00\u4e0b\u6211\u4eec\u8981\u6784\u5efa\u7684\u5185\u5bb9：<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120304-688a09f8dbeae.webp\")
<\/p>\n
\u6211\u4eec\u7684 MCP \u670d\u52a1\u5668\u5145\u5f53 LLM Agent \u548c\u4f60\u7684 Elasticsearch \u6570\u636e\u4e4b\u95f4\u7684\u6865\u6881，\u4f7f\u4f60\u80fd\u591f\u5bf9\u7ed3\u6784\u5316\u5065\u5eb7\u6570\u636e\u8fdb\u884c\u81ea\u7136\u8bed\u8a00\u67e5\u8be2\u3002<\/p>\n<\/p>\n
\u672c\u6587\u5047\u8bbe\u4f60\u5df2\u5b89\u88c5\u4ee5\u4e0b\u6280\u672f：<\/p>\n
Python 3.10+<\/p>\n<\/li>\n
Elasticsearch（\u5feb\u901f\u5b89\u88c5\u6307\u5357）<\/p>\n<\/li>\n
Claude Desktop（\u514d\u8d39\u4e0b\u8f7d）<\/p>\n<\/li>\n
Python \u7684 uv \u5305\u7ba1\u7406\u5668（\u89c1\u5b89\u88c5\u8bf4\u660e）<\/p>\n<\/li>\n<\/ul>\n
\u6574\u4e2a\u9879\u76ee\u5305\u542b\u811a\u672c\u548c\u793a\u4f8b\u6570\u636e，\u5747\u6258\u7ba1\u5728\u8fd9\u4e2a GitHub \u4ed3\u5e93\u4e2d\u3002\u5982\u679c\u4f60\u613f\u610f，\u53ef\u4ee5\u6309\u7167 README.md \u6587\u4ef6\u4e2d\u7684\u8bf4\u660e\u5feb\u901f\u5b8c\u6210\u73af\u5883\u642d\u5efa\u3002<\/p>\n
\u5728\u7ec8\u7aef\u4e2d\u514b\u9686\u8be5\u4ed3\u5e93，\u4f7f\u7528\u8f85\u52a9\u811a\u672c\u5982 ingest_data.py \u6765\u51c6\u5907\u6570\u636e\u5e93，\u7136\u540e\u518d\u7ee7\u7eed\u540e\u7eed\u7ae0\u8282\u3002<\/p>\n
git clone https:\/\/github.com\/salgado\/apple-watch-health-mcp.git
\ncd apple-watch-health-mcp <\/p>\n
\u5f53\u4f60\u5217\u51fa\u6587\u4ef6\u5939\u5185\u5bb9（\u4f7f\u7528 ls \u547d\u4ee4）\u65f6，\u4f60\u5e94\u8be5\u770b\u5230\u4ee5\u4e0b\u6587\u4ef6：<\/p>\n
apple_watch_mcp.py：MCP \u670d\u52a1\u5668\u7684\u5b8c\u6574\u811a\u672c，\u5305\u542b\u6240\u6709\u5b9e\u73b0\u7684 Resources\u3001Tools \u548c Prompts\u3002<\/p>\n<\/li>\n
ingest_data.py：\u8f85\u52a9\u811a\u672c，\u7528\u4e8e\u5c06\u793a\u4f8b\u6570\u636e\u5bfc\u5165\u5230\u4f60\u7684 Elasticsearch \u5b9e\u4f8b\u3002<\/p>\n<\/li>\n
sample_data.json：\u5305\u542b\u865a\u6784 Apple Health \u6b65\u6570\u6570\u636e\u7684\u793a\u4f8b JSON \u6587\u4ef6\u3002<\/p>\n<\/li>\n
README.md：\u5b8c\u6574\u89e3\u51b3\u65b9\u6848\u7684\u8fd0\u884c\u8bf4\u660e\u3002<\/p>\n<\/li>\n<\/ul>\n
\u73b0\u5728，\u8ba9\u6211\u4eec\u4e3a MCP \u670d\u52a1\u5668\u642d\u5efa\u73af\u5883\u3002FastMCP \u662f\u4e00\u4e2a Python \u6846\u67b6，\u53ef\u4ee5\u8f7b\u677e\u521b\u5efa\u4e0e MCP \u534f\u8bae\u517c\u5bb9\u7684\u5b8c\u6574\u670d\u52a1\u5668\u3002<\/p>\n<\/p>\n
\u5728\u6784\u5efa\u81ea\u5b9a\u4e49\u670d\u52a1\u5668\u4e4b\u524d，\u4e86\u89e3\u4e24\u79cd\u96c6\u6210 MCP \u4e0e Elasticsearch \u7684\u65b9\u5f0f\u5f88\u91cd\u8981：<\/p>\n<\/p>\n
1) \u5b98\u65b9\u65b9\u6848：\u4f7f\u7528 Elastic MCP Server<\/p>\n
\u901a\u7528：\u9002\u7528\u4e8e\u5927\u591a\u6570\u7528\u4f8b<\/p>\n<\/li>\n
\u914d\u7f6e\u7b80\u5355：\u5feb\u901f\u8f7b\u677e\u642d\u5efa<\/p>\n<\/li>\n
\u5b9a\u5236\u6709\u9650：\u7279\u5b9a\u573a\u666f\u7075\u6d3b\u6027\u8f83\u4f4e<\/p>\n<\/li>\n
\u64cd\u4f5c\u53d7\u9650：\u64b0\u5199\u65f6\u4ec5\u652f\u6301\u57fa\u672c\u67e5\u8be2\u64cd\u4f5c<\/p>\n<\/li>\n
\u535a\u5ba2：\u5c06\u4ee3\u7406\u8fde\u63a5\u5230 Elasticsearch \u4f7f\u7528\u6a21\u578b\u4e0a\u4e0b\u6587\u534f\u8bae<\/p>\n<\/li>\n<\/ul>\n
2) \u81ea\u5b9a\u4e49\u65b9\u6848：\u6784\u5efa\u4f60\u81ea\u5df1\u7684 MCP \u670d\u52a1\u5668（\u672c\u6587\u91cd\u70b9）<\/p>\n
\u9886\u57df\u4e13\u7528：\u9488\u5bf9\u7279\u5b9a\u7528\u4f8b\u4f18\u5316<\/p>\n<\/li>\n
\u9700\u8981\u7f16\u7a0b：\u5f00\u53d1\u5de5\u4f5c\u91cf\u8f83\u5927<\/p>\n<\/li>\n
\u5b8c\u5168\u63a7\u5236：\u529f\u80fd\u7075\u6d3b\u53ef\u63a7<\/p>\n<\/li>\n<\/ul>\n
\u9996\u5148，\u8fdb\u5165\u9879\u76ee\u6587\u4ef6\u5939，\u4f7f\u7528 uv \u7ba1\u7406\u4f9d\u8d56：<\/p>\n
# Navigate to the repository folder
\ncd apple-watch-health-mcp<\/p>\n
# Create and initialize a new project
\nuv init <\/p>\n
# Create and activate a virtual environment
\nuv venv
\nsource .venv\/bin\/activate <\/p>\n
# Install required dependencies
\nuv add "mcp[cli]" "elasticsearch>=8.0.0,<9.0.0" aiohttp pydantic<\/p>\n
# Set up environment variable for API key (after creating it in Elasticsearch)
\nexport ES_API_KEY="your_encoded_api_key_here" <\/p>\n
\u6ce8\u610f，\u6211\u4eec\u6307\u5b9a\u4e86\u4e0e\u672c\u535a\u5ba2\u4e2d\u4f7f\u7528\u7684 Elasticsearch v8.x \u670d\u52a1\u5668\u517c\u5bb9\u7684\u5e93\u7248\u672c\u3002<\/p>\n<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120305-688a09f9230d3.webp\")
<\/p>\n
\u56fe\u7247\u6765\u6e90\u6539\u7f16\u81ea Norah Sakal：
\n What is Model Context Protocol (MCP)? How it simplifies AI integrations compared to APIs | AI Agents That Work<\/p>\n
\u8981\u83b7\u53d6 Apple HealthKit \u6570\u636e，\u8bf7\u53c2\u9605 Apple \u5b98\u65b9\u5bfc\u51fa\u6307\u5357\u3002\u5bfc\u51fa\u4f1a\u751f\u6210\u5305\u542b\u6240\u6709\u5065\u5eb7\u8bb0\u5f55\u7684 XML \u6587\u4ef6\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120305-688a09f94a99a.webp\")
<\/p>\n<\/p>\n
\u5728\u521b\u5efa MCP \u670d\u52a1\u5668\u4e4b\u524d，\u6211\u4eec\u9700\u8981\u7528\u793a\u4f8b\u6570\u636e\u51c6\u5907 Elasticsearch\u3002\u4ed3\u5e93\u4e2d\u5df2\u5305\u542b\u4e00\u4e2a\u82f9\u679c\u5065\u5eb7\u6b65\u6570\u6570\u636e\u6587\u4ef6（sample_data.json）\u548c\u4e00\u4e2a\u5bfc\u5165\u811a\u672c（ingest_data.py）\u3002 \u8981\u51c6\u5907 Elasticsearch，\u8bf7\u5728\u7ec8\u7aef\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4：<\/p>\n
# Set the environment variable with Elasticsearch API KEY
\nexport ES_API_KEY="ES_API_KEY"
\npython ingest_data.py <\/p>\n
\u8be5\u811a\u672c\u5c06\u4f1a：<\/p>\n
\u8fde\u63a5\u5230 http:\/\/localhost:9200 \u7684 Elasticsearch<\/p>\n<\/li>\n
\u4f7f\u7528\u6b63\u786e\u7684\u6620\u5c04\u521b\u5efa\u7d22\u5f15<\/p>\n<\/li>\n
\u4ece JSON \u6587\u4ef6\u63d2\u5165\u6570\u636e<\/p>\n<\/li>\n<\/ul>\n
\u5982\u679c\u4f60\u60f3\u4f7f\u7528\u5176\u4ed6\u8fde\u63a5\u9009\u9879，\u6bd4\u5982 Elastic Cloud，\u8bf7\u67e5\u9605\u76f8\u5173\u6587\u6863\u3002\u9ed8\u8ba4\u914d\u7f6e\u4f7f\u7528：<\/p>\n
ES_HOST = "http:\/\/localhost:9200"
\nES_INDEX = "apple-health-steps"
\nES_API_KEY = os.getenv('ES_API_KEY') <\/p>\n
\u8fc7\u7a0b\u7ed3\u675f\u65f6\u4f60\u5e94\u8be5\u4f1a\u770b\u5230\u6210\u529f\u6d88\u606f\u3002<\/p>\n
Connecting to Elasticsearch at http:\/\/localhost:9200…
\nConnection successful!
\nIndex 'apple-health-steps' found. Deleting…
\nIndex deleted.
\nCreating index 'apple-health-steps' with the specified mapping…
\nIndex created successfully.
\nReading data from 'sample_data.json' for ingestion…
\nIngestion complete. Documents successfully ingested: 68
\nFinal check: The index 'apple-health-steps' now contains 68 documents. <\/p>\n
\u811a\u672c\u6267\u884c\u5b8c\u6210\u540e，\u7d22\u5f15\u5c06\u88ab\u521b\u5efa\u5e76\u586b\u5145\u793a\u4f8b\u6570\u636e\u3002\u8be5\u7d22\u5f15\u9075\u5faa\u4ee5\u4e0b\u6a21\u5f0f，\u5bf9\u5e94\u4ece Apple Watch \u5bfc\u51fa\u7684\u6b65\u6570\u6570\u636e XML \u6587\u4ef6\u4e2d\u7684\u5b57\u6bb5：<\/p>\n
GET \/apple-health-steps\/_mapping
\n{
\n "apple-health-steps": {
\n "mappings": {
\n "properties": {
\n "creationDate": { "type": "date", "format": "yyyy-MM-dd HH:mm:ss||yyyy-MM-dd||epoch_millis" },
\n "day": { "type": "date", "format": "yyyy-MM-dd" },
\n "dayOfWeek": { "type": "keyword" },
\n "deviceInfo": {
\n "properties": {
\n "hardware": { "type": "keyword" },
\n "manufacturer": { "type": "keyword" },
\n "model": { "type": "keyword" },
\n "name": { "type": "keyword" },
\n "software": { "type": "keyword" }
\n }
\n },
\n "duration": { "type": "float" },
\n "endDate": { "type": "date", "format": "yyyy-MM-dd HH:mm:ss||yyyy-MM-dd||epoch_millis" },
\n "hour": { "type": "integer" },
\n "sourceName": { "type": "keyword" },
\n "startDate": { "type": "date", "format": "yyyy-MM-dd HH:mm:ss||yyyy-MM-dd||epoch_millis" },
\n "type": { "type": "keyword" },
\n "unit": { "type": "keyword" },
\n "value": { "type": "float" }
\n }
\n }
\n }
\n} <\/p>\n
\u4ee5\u4e0b\u662f\u63d2\u5165\u7684\u6570\u636e\u793a\u4f8b：<\/p>\n
{
\n "type": "HKQuantityTypeIdentifierStepCount",
\n "sourceName": "Apple Watch",
\n "deviceInfo": {
\n "name": "Apple Watch",
\n "manufacturer": "Apple Inc.",
\n "model": "Watch",
\n "hardware": "Watch7,1",
\n "software": "11.3.1"
\n },
\n "unit": "count",
\n "creationDate": "2025-06-01 08:11:04",
\n "startDate": "2025-06-01 07:58:42",
\n "endDate": "2025-06-01 07:59:10",
\n "value": 340,
\n "day": "2025-06-01",
\n "dayOfWeek": "Sunday",
\n "hour": 8,
\n "duration": 583
\n} <\/p>\n
\u6570\u636e\u5305\u542b\u6765\u81ea Apple Watch \u548c iPhone \u4e24\u4e2a\u6765\u6e90\u7684\u6b65\u6570\u8ba1\u6570，\u8bb0\u5f55\u4e86\u8be6\u7ec6\u7684\u65f6\u95f4\u6233\u548c\u8bbe\u5907\u4fe1\u606f\u3002<\/p>\n
\u6570\u636e\u5e93\u521b\u5efa\u5e76\u586b\u5145\u5b8c\u6210\u540e，\u6211\u4eec\u53ef\u4ee5\u5f00\u59cb\u6784\u5efa MCP \u670d\u52a1\u5668\u6765\u8bbf\u95ee\u8fd9\u4e9b\u6570\u636e\u3002<\/p>\n<\/p>\n
\u8ba9\u6211\u4eec\u521b\u5efa\u4e00\u4e2a apple_watch_mcp.py \u6587\u4ef6，\u5305\u542b MCP \u670d\u52a1\u5668\u7684\u57fa\u672c\u7ed3\u6784：<\/p>\n
# apple_watch_health_mcp.py
\nimport os
\nfrom typing import Any, Optional
\nimport json
\nfrom datetime import datetime
\nfrom pydantic import BaseModel, field_validator, ValidationError
\nfrom mcp.server.fastmcp import FastMCP
\nfrom elasticsearch import AsyncElasticsearch
\nfrom contextlib import asynccontextmanager<\/p>\n
# Initialize FastMCP server
\nmcp = FastMCP("apple-watch-steps")<\/p>\n
# Constants
\nES_HOST = "http:\/\/localhost:9200"
\nES_INDEX = "apple-health-steps"<\/p>\n
# API key from environment variable or fallback for development
\nES_API_KEY = os.getenv('ES_API_KEY')<\/p>\n
# Pydantic model for parameter validation
\nclass QueryStepDataParams(BaseModel):
\n start_date: Optional[str] = None
\n end_date: Optional[str] = None
\n aggregation: Optional[str] = None
\n device: Optional[str] = None<\/p>\n
@field_validator('start_date', 'end_date')
\n def validate_date_format(cls, value):
\n if value is None:
\n return value
\n try:
\n datetime.strptime(value, "%Y-%m-%d")
\n return value
\n except ValueError:
\n raise ValueError("Invalid date format. Use YYYY-MM-DD")<\/p>\n
@field_validator('aggregation')
\n def validate_aggregation(cls, value):
\n valid_aggregations = ["hourly", "daily", "weekly", "monthly", None]
\n if value not in valid_aggregations:
\n raise ValueError(f"Invalid aggregation. Use one of: {valid_aggregations[:-1]}")
\n return value<\/p>\n
# Main function to run the server
\nif __name__ == "__main__":
\n mcp.run() <\/p>\n
\u8fd9\u6bb5\u521d\u59cb\u4ee3\u7801\u8bbe\u7f6e\u4e86\u4e00\u4e2a\u540d\u4e3a \u201capple-watch-steps\u201d \u7684 MCP \u670d\u52a1\u5668，\u5e76\u5b9a\u4e49\u4e86\u4e00\u4e2a Pydantic \u6a21\u578b\u6765\u9a8c\u8bc1\u5de5\u5177\u4e2d\u4f7f\u7528\u7684\u53c2\u6570\u3002\u670d\u52a1\u5668\u5c06\u8fde\u63a5\u5230\u672c\u5730 Elasticsearch \u96c6\u7fa4，\u4f7f\u7528 \u201capple-health-steps\u201d \u7d22\u5f15\u3002<\/p>\n<\/p>\n
\u4e3a\u4e86\u4e0e Elasticsearch \u4ea4\u4e92，\u8ba9\u6211\u4eec\u5728\u4ee3\u7801\u4e2d\u6dfb\u52a0\u4e00\u4e2a\u8f85\u52a9\u51fd\u6570：<\/p>\n
@asynccontextmanager
\nasync def get_es_client():
\n """Context manager for Elasticsearch client."""
\n client = AsyncElasticsearch([ES_HOST], api_key=ES_API_KEY)
\n try:
\n yield client
\n finally:
\n await client.close()<\/p>\n
# Elasticsearch helper function
\nasync def query_elasticsearch(query: dict) -> dict[str, Any] | None:
\n """Makes a request to Elasticsearch with proper error handling."""
\n print(f"Sending query to Elasticsearch: {json.dumps(query)}")<\/p>\n
# Use context manager
\n async with get_es_client() as client:
\n try:
\n response = await client.search(
\n index=ES_INDEX,
\n body=query
\n )
\n return response
\n except Exception as e:
\n print(f"Error querying Elasticsearch: {e}")
\n return None <\/p>\n
\u8be5\u5b9e\u73b0\u4f7f\u7528\u5b98\u65b9 Elasticsearch Python \u5ba2\u6237\u7aef\u4e2d\u7684 AsyncElasticsearch \u7c7b，\u652f\u6301\u5b8c\u6574\u7684 async\/await \u5f02\u6b65\u64cd\u4f5c\u3002\u91c7\u7528\u5f02\u6b65\u65b9\u5f0f\u66ff\u4ee3\u4f20\u7edf\u540c\u6b65\u5ba2\u6237\u7aef\u64cd\u4f5c，\u53ef\u4ee5\u907f\u514d\u957f\u65f6\u95f4\u4efb\u52a1\u963b\u585e\u4e3b\u7ebf\u7a0b，\u4f7f\u670d\u52a1\u5668\u54cd\u5e94\u66f4\u5feb\u3002\u4e0a\u4e0b\u6587\u7ba1\u7406\u5668\u6a21\u5f0f（context manager pattern）\u786e\u4fdd\u6bcf\u6b21\u64cd\u4f5c\u65f6\u8fde\u63a5\u6b63\u786e\u6253\u5f00\u548c\u5173\u95ed，\u9632\u6b62\u8d44\u6e90\u6cc4\u6f0f\u548c\u8fde\u63a5\u8b66\u544a\u3002\u51fd\u6570\u63a5\u6536\u4e00\u4e2a\u67e5\u8be2\u5b57\u5178，\u53d1\u9001\u5230 Elasticsearch，\u5e76\u8fd4\u56de JSON \u683c\u5f0f\u7ed3\u679c；\u5982\u679c\u51fa\u9519\u5219\u8fd4\u56de None\u3002<\/p>\n<\/p>\n
\u73b0\u5728\u6211\u4eec\u6765\u5b9e\u73b0 MCP \u7684 Resources\u3001Tools \u548c Prompts，\u6bcf\u4e2a\u7ec4\u4ef6\u6709\u5176\u7279\u5b9a\u89d2\u8272\u3002<\/p>\n
\u793a\u4f8b Resource：<\/p>\n
@mcp.resource("health:\/\/steps\/latest")
\nasync def get_latest_steps() -> str:
\n """Gets the most recent step records"""
\n query = {
\n "query": {
\n "match_all": {}
\n },
\n "sort": [
\n {"endDate": {"order": "desc"}}
\n ],
\n "size": 10
\n }<\/p>\n
data = await query_elasticsearch(query)
\n if not data:
\n return json.dumps({"error": "Unable to query Elasticsearch"}, indent=2)<\/p>\n
results = []
\n for hit in data["hits"]["hits"]:
\n source = hit["_source"]
\n results.append({
\n "startDate": source.get("startDate"),
\n "endDate": source.get("endDate"),
\n "value": source.get("value"),
\n "device": source.get("device"),
\n "sourceName": source.get("sourceName"),
\n "dayOfWeek": source.get("dayOfWeek"),
\n "hour": source.get("hour")
\n })<\/p>\n
return json.dumps({
\n "latest_steps": results
\n }, indent=2) <\/p>\n
\u8be5 Resource \u4f1a\u67e5\u8be2 Elasticsearch，\u83b7\u53d6\u6309\u7ed3\u675f\u65f6\u95f4\u6392\u5e8f\u7684\u6700\u8fd1 10 \u6761\u6b65\u6570\u8bb0\u5f55\u3002\u672c\u6587\u540e\u9762\u4f1a\u901a\u8fc7\u5b9e\u9645 Claude Desktop \u4ea4\u4e92\u793a\u4f8b\u5c55\u793a\u8f93\u51fa\u683c\u5f0f\u548c\u6570\u636e\u6548\u679c\u3002<\/p>\n<\/p>\n
\u6211\u4eec\u7684\u7528\u4f8b\u662f\u521b\u5efa\u4e00\u4e2a\u77e5\u8bc6\u5e93\u52a9\u624b，\u4f7f LLM Agent \u80fd\u4e0e Apple HealthKit \u5065\u5eb7\u6570\u636e\u4ea4\u4e92\u3002\u91cd\u70b9\u662f\u6b65\u6570\u6570\u636e，\u4f46\u540c\u6837\u65b9\u6cd5\u53ef\u6269\u5c55\u5230\u5176\u4ed6\u5065\u5eb7\u6570\u636e\u7c7b\u578b\u3002<\/p>\n
\u8be5\u52a9\u624b\u5141\u8bb8\u7528\u6237：<\/p>\n
\u7528\u5bf9\u8bdd\u65b9\u5f0f\u67e5\u8be2\u6b65\u6570\u6570\u636e<\/p>\n<\/li>\n
\u83b7\u53d6\u4e2a\u6027\u5316\u5206\u6790\u548c\u53ef\u89c6\u5316<\/p>\n<\/li>\n
\u6bd4\u8f83\u4e0d\u540c\u8bbe\u5907\u7684\u6570\u636e<\/p>\n<\/li>\n
\u8bc6\u522b\u957f\u671f\u8d8b\u52bf\u548c\u6a21\u5f0f<\/p>\n<\/li>\n<\/ul>\n
\u6211\u4eec\u7684 MCP \u670d\u52a1\u5668\u5c06\u5b9e\u73b0\u4e09\u4e2a\u4e3b\u8981 Resource：<\/p>\n
\u6570\u636e\u7c7b\u578b\u5217\u8868：\u663e\u793a\u6570\u636e\u5e93\u4e2d\u53ef\u7528\u7684\u6570\u636e\u7c7b\u578b<\/p>\n<\/li>\n
\u6700\u8fd1\u6570\u636e\u83b7\u53d6：\u83b7\u53d6\u6700\u65b0\u7684\u6b65\u6570\u8bb0\u5f55<\/p>\n<\/li>\n
\u7edf\u8ba1\u6c47\u603b：\u63d0\u4f9b\u6570\u636e\u6982\u89c8<\/p>\n<\/li>\n<\/ul>\n
\u8ba9\u6211\u4eec\u5f00\u59cb\u5b9e\u73b0\u8fd9\u4e9b Resources：<\/p>\n
# Resources
\n@mcp.resource("health:\/\/steps\/types")
\nasync def list_step_types() -> str:
\n """List all available step types in the database"""
\n query = {
\n "size": 0,
\n "aggs": {
\n "step_types": {
\n "terms": {
\n "field": "type",
\n "size": 100
\n }
\n }
\n }
\n }<\/p>\n
data = await query_elasticsearch(query)
\n if not data:
\n return json.dumps({"error": "Unable to query Elasticsearch"}, indent=2)<\/p>\n
step_types = [bucket["key"] for bucket in
\n data["aggregations"]["step_types"]["buckets"]]<\/p>\n
return json.dumps({
\n "available_types": step_types,
\n "count": len(step_types)
\n }, indent=2)<\/p>\n
@mcp.resource("health:\/\/steps\/summary")
\nasync def get_steps_summary() -> str:
\n """Get summary statistics for step counts"""
\n query = {
\n "aggs": {
\n "all_time": {
\n "stats": {
\n "field": "value"
\n }
\n }
\n },
\n "size": 0
\n }<\/p>\n
data = await query_elasticsearch(query)
\n if not data:
\n return json.dumps({"error": "Unable to query Elasticsearch"}, indent=2)<\/p>\n
return json.dumps(data["aggregations"], indent=2) <\/p>\n
\u4e3a\u4e86\u5c06\u6211\u4eec\u7684 MCP \u670d\u52a1\u5668\u4e0e RAG（\u68c0\u7d22\u589e\u5f3a\u751f\u6210）\u5de5\u4f5c\u6d41\u7a0b\u96c6\u6210，\u6211\u4eec\u9700\u8981\u5b9e\u73b0\u652f\u6301\u81ea\u5b9a\u4e49\u590d\u6742\u67e5\u8be2\u7684\u5de5\u5177\u3002\u4e0b\u56fe\u5c55\u793a\u4e86 MCP \u5982\u4f55\u878d\u5165\u5b8c\u6574\u7684 RAG \u6d41\u7a0b：<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120305-688a09f985b60.webp\")
<\/p>\n
\u8be5\u56fe\u5c55\u793a\u4e86\u6211\u4eec\u7684 MCP \u670d\u52a1\u5668\u5982\u4f55\u5b8c\u7f8e\u878d\u5165 RAG \u6d41\u7a0b：<\/p>\n
\u7528\u6237\u63d0\u51fa\u5173\u4e8e\u5065\u5eb7\u6570\u636e\u7684\u95ee\u9898<\/p>\n<\/li>\n
\u67e5\u8be2\u88ab RAG \u7cfb\u7edf\u5904\u7406<\/p>\n<\/li>\n
\u68c0\u7d22\u9636\u6bb5\u4f7f\u7528 MCP \u7684 query_step_data \u5de5\u5177\u5728 Elasticsearch \u4e2d\u641c\u7d22\u76f8\u5173\u6570\u636e<\/p>\n<\/li>\n
\u68c0\u7d22\u5230\u7684\u6570\u636e\u7528\u4e8e\u6784\u5efa\u4e0a\u4e0b\u6587\u5e76\u4e30\u5bcc\u56de\u590d\u5185\u5bb9<\/p>\n<\/li>\n
LLM \u6839\u636e\u6570\u636e\u548c\u4e0a\u4e0b\u6587\u751f\u6210\u6700\u7ec8\u56de\u590d<\/p>\n<\/li>\n<\/ul>\n
\u8ba9\u6211\u4eec\u5b9e\u73b0\u8fd9\u4e2a\u6d41\u7a0b\u6240\u9700\u7684\u5de5\u5177：<\/p>\n
# Tools
\n@mcp.tool()
\nasync def query_step_data(params: QueryStepDataParams) -> str:
\n """
\n Query step data with customizable parameters<\/p>\n
Args:
\n start_date: Start date in YYYY-MM-DD format
\n end_date: End date in YYYY-MM-DD format
\n aggregation: Aggregation interval (hourly, daily, weekly, monthly)
\n device: Filter by specific device name
\n """
\n # Extract parameters from model
\n start_date = params.start_date or ""
\n end_date = params.end_date or ""
\n aggregation = params.aggregation or ""
\n device = params.device or ""<\/p>\n
query = {"query": {"match_all": {}}}
\n filters = []<\/p>\n
# Date filters
\n date_ranges = []
\n if start_date:
\n date_ranges.append({"gte": start_date})
\n if end_date:
\n date_ranges.append({"lte": end_date})<\/p>\n
if date_ranges:
\n filters.append({
\n "range": {
\n "day": {**{k: v for d in date_ranges for k, v in d.items()}}
\n }
\n })<\/p>\n
# Device filter
\n if device:
\n filters.append({
\n "wildcard": {
\n "device": f"*{device}*"
\n }
\n })<\/p>\n
if filters:
\n query["query"] = {"bool": {"must": filters}}<\/p>\n
# Aggregation handling
\n if aggregation:
\n interval_mapping = {
\n "hourly": "1h",
\n "daily": "1d",
\n "weekly": "1w",
\n "monthly": "1M"
\n }
\n interval = interval_mapping.get(aggregation, "1d")
\n date_field = "startDate" if aggregation == "hourly" else "day"<\/p>\n
query["aggs"] = {
\n "time_series": {
\n "date_histogram": {
\n "field": date_field,
\n "calendar_interval": interval,
\n "min_doc_count": 0
\n },
\n "aggs": {
\n "total_steps": {"sum": {"field": "value"}},
\n "avg_steps": {"avg": {"field": "value"}},
\n "max_steps": {"max": {"field": "value"}},
\n "min_steps": {"min": {"field": "value"}}
\n }
\n }
\n }
\n query["size"] = 0
\n else:
\n query.update({
\n "sort": [{"startDate": "desc"}],
\n "size": 10
\n })<\/p>\n
data = await query_elasticsearch(query)
\n if not data:
\n return json.dumps({
\n "error": "Unable to query Elasticsearch",
\n "query": query
\n }, indent=2)<\/p>\n
# Process results
\n results = []
\n if aggregation and "time_series" in data.get("aggregations", {}):
\n for bucket in data["aggregations"]["time_series"]["buckets"]:
\n results.append({
\n "date": bucket["key_as_string"],
\n "total_steps": bucket["total_steps"]["value"],
\n "average_steps": bucket["avg_steps"]["value"],
\n "max_steps": bucket["max_steps"]["value"],
\n "min_steps": bucket["min_steps"]["value"],
\n "records": bucket["doc_count"]
\n })
\n else:
\n for hit in data["hits"]["hits"]:
\n source = hit["_source"]
\n results.append({
\n "startDate": source.get("startDate"),
\n "endDate": source.get("endDate"),
\n "day": source.get("day"),
\n "dayOfWeek": source.get("dayOfWeek"),
\n "hour": source.get("hour"),
\n "value": source.get("value"),
\n "device": source.get("device"),
\n "sourceName": source.get("sourceName")
\n })<\/p>\n
return json.dumps({
\n "aggregation": aggregation,
\n "total_records": data["hits"]["total"]["value"],
\n "data": results,
\n "query": query
\n }, indent=2)<\/p>\n
@mcp.tool()
\nasync def get_all_steps() -> str:
\n """Get all steps without any filtering"""
\n query = {
\n "query": {
\n "match_all": {}
\n },
\n "size": 10,
\n "sort": [{"startDate": "desc"}]
\n }<\/p>\n
data = await query_elasticsearch(query)
\n if not data:
\n return json.dumps({"error": "Unable to query Elasticsearch"}, indent=2)<\/p>\n
results = []
\n for hit in data["hits"]["hits"]:
\n source = hit["_source"]
\n results.append({
\n "startDate": source.get("startDate"),
\n "value": source.get("value"),
\n "device": source.get("device")
\n })<\/p>\n
return json.dumps({
\n "total_records": data["hits"]["total"]["value"],
\n "data": results
\n }, indent=2) <\/p>\n
\u8fd9\u4e9b\u5de5\u5177\u652f\u6301：<\/p>\n
\u5e26\u65e5\u671f\u3001\u8bbe\u5907\u548c\u805a\u5408\u8fc7\u6ee4\u7684\u81ea\u5b9a\u4e49\u67e5\u8be2<\/p>\n<\/li>\n
\u65e0\u8fc7\u6ee4\u6761\u4ef6\u7684\u5feb\u901f\u6700\u8fd1\u6570\u636e\u68c0\u7d22<\/p>\n<\/li>\n<\/ul>\n
\u6211\u4eec\u7684 MCP \u670d\u52a1\u5668\u53ef\u4e0e\u4e0d\u540c LLM \u670d\u52a1\u5546\u914d\u5408\u4f7f\u7528\u3002\u4e3a\u4e86\u65b9\u4fbf\u96c6\u6210，\u8ba9\u6211\u4eec\u5b9e\u73b0\u4e00\u4e9b prompts，\u5e2e\u52a9 LLM \u6709\u6548\u5229\u7528\u6211\u4eec\u7684\u8d44\u6e90\u548c\u5de5\u5177：<\/p>\n
@mcp.prompt()
\ndef daily_report(date: str = None) -> str:
\n """Create a daily step report for a specific date"""
\n if date:
\n return f"""Please analyze the step data for {date}.
\nProvide:
\n1. Total steps
\n2. Average steps per active hour
\n3. Most active periods of the day
\n4. Comparison with weekly average
\n5. Graphical visualization of the data, if possible"""
\n else:
\n return """Please analyze the step data for today.
\nProvide:
\n1. Total steps so far
\n2. Average steps per active hour
\n3. Most active periods of the day
\n4. Comparison with weekly average
\n5. Graphical visualization of the data, if possible"""<\/p>\n
@mcp.prompt()
\ndef trend_analysis(start_date: str, end_date: str) -> str:
\n """Analyze step trends over a specific period"""
\n return f"""Analyze step trends between {start_date} and {end_date}.
\nPlease include:
\n1. Daily step trend graph
\n2. Identification of weekly patterns
\n3. Days with highest and lowest activity
\n4. Progression over time
\n5. Recommendations based on the data"""<\/p>\n
@mcp.prompt()
\ndef device_comparison() -> str:
\n """Compare step data recorded by different devices"""
\n return """Compare step data recorded by Apple Watch and iPhone:
\n1. Which device records more steps on average?
\n2. Are there notable differences in usage patterns?
\n3. Times when each device is used more
\n4. Apparent accuracy of each device
\n5. Recommendations on which device to prioritize for tracking""" <\/p>\n
\u5728\u5c06\u670d\u52a1\u5668\u8fde\u63a5\u5230\u771f\u5b9e\u7684 LLM \u5ba2\u6237\u7aef（\u6bd4\u5982 Claude）\u4e4b\u524d，\u6700\u597d\u5148\u5728\u53d7\u63a7\u73af\u5883\u4e2d\u8fdb\u884c\u6d4b\u8bd5\u3002mcp \u5e93\u81ea\u5e26\u4e00\u4e2a\u5f3a\u5927\u7684\u57fa\u4e8e\u7f51\u9875\u7684\u8c03\u8bd5\u5de5\u5177，\u53eb\u505a MCP Inspector\u3002\u5b83\u5141\u8bb8\u6211\u4eec\u4ea4\u4e92\u5f0f\u6d4b\u8bd5\u670d\u52a1\u5668，\u5b9e\u65f6\u67e5\u770b\u8bf7\u6c42\u548c\u54cd\u5e94，\u9a8c\u8bc1\u903b\u8f91，\u786e\u4fdd\u90e8\u7f72\u524d\u65e0\u8bef\u3002<\/p>\n<\/p>\n
\u5728\u6fc0\u6d3b\u865a\u62df\u73af\u5883\u7684\u7ec8\u7aef\u4e2d，\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4，\u540e\u63a5\u4f60\u7684\u811a\u672c\u540d\u79f0：<\/p>\n
mcp dev apple_watch_mcp.py <\/p>\n
\u4f60\u7b2c\u4e00\u6b21\u8fd0\u884c\u65f6，mcp \u53ef\u80fd\u4f1a\u8bf7\u6c42\u5b89\u88c5 Inspector \u7684\u7f51\u9875\u7ec4\u4ef6\u3002\u8fd9\u662f\u4e00\u6b21\u6027\u8bbe\u7f6e\u3002\u53ea\u9700\u8f93\u5165 y \u7136\u540e\u6309\u56de\u8f66\u5373\u53ef\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120305-688a09f9ae895.webp\")
<\/p>\n
\u786e\u8ba4\u540e，\u8be5\u5de5\u5177\u4f1a\u542f\u52a8\u672c\u5730\u4ee3\u7406\u670d\u52a1\u5668，\u5e76\u81ea\u52a8\u5728\u9ed8\u8ba4\u6d4f\u89c8\u5668\u6253\u5f00 MCP Inspector \u754c\u9762，\u901a\u5e38\u5730\u5740\u662f http:\/\/127.0.0.1:6274\u3002<\/p>\n
\u8ba4\u8bc1\u8bf4\u660e：\u65b0\u7248 MCP Inspector \u53ef\u80fd\u9700\u8981\u8ba4\u8bc1\u4ee5\u4fdd\u969c\u5b89\u5168\u3002\u5982\u679c\u754c\u9762\u672a\u81ea\u52a8\u6253\u5f00，\u8bf7\u67e5\u770b\u7ec8\u7aef\u4e2d\u7684\u63d0\u793a\u4fe1\u606f，\u5982：<\/p>\n
Session token: xxxxx…
\nOpen inspector with token pre-filled:
\n http:\/\/localhost:6274\/?MCP_PROXY_AUTH_TOKEN=xxxxx… <\/p>\n
\u4f7f\u7528\u5e26\u6709\u4ee4\u724c\u7684\u5b8c\u6574 URL \u8bbf\u95ee Inspector\u3002\u8be5\u4ee4\u724c\u786e\u4fdd\u6d4b\u8bd5\u671f\u95f4\u5bf9\u670d\u52a1\u5668\u7684\u5b89\u5168\u8bbf\u95ee\u3002<\/p>\n<\/p>\n
MCP Inspector \u63d0\u4f9b\u4e86\u7b80\u6d01\u7684\u754c\u9762，\u65b9\u4fbf\u4f60\u63a2\u7d22\u548c\u4ea4\u4e92\u670d\u52a1\u5668\u529f\u80fd\u3002<\/p>\n<\/p>\n
\u9875\u9762\u52a0\u8f7d\u65f6，\u4f1a\u81ea\u52a8\u8fde\u63a5\u5230\u4f60\u8fd0\u884c\u7684 apple_watch_mcp.py \u811a\u672c\u3002\u5de6\u4fa7\u72b6\u6001\u4f1a\u4ece\u672a\u8fde\u63a5\u53d8\u4e3a\u201cRestart | Disconnect\u201d，\u5e76\u4e14 \u201cHistory\u201d \u9762\u677f\u4f1a\u663e\u793a\u4e00\u4e2a initialize \u4e8b\u4ef6\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120305-688a09f9cbaa6.webp\")
<\/p>\n<\/p>\n
Inspector \u7684\u6838\u5fc3\u5728\u9876\u90e8\u7684\u6807\u7b7e\u9875：Resources\u3001Prompts \u548c Tools\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120305-688a09f9f228b.webp\")
<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120306-688a09fa22c8b.webp\")
<\/p>\n<\/p>\n
\u8ba9\u6211\u4eec\u6d4b\u8bd5\u4e00\u4e2a\u7b80\u5355\u7684\u5de5\u5177，\u770b\u770b\u5b83\u7684\u8fd0\u884c\u6548\u679c\u3002<\/p>\n
\u8fdb\u5165 Tools \u6807\u7b7e\u9875<\/p>\n<\/li>\n
\u4ece\u5217\u8868\u4e2d\u70b9\u51fb\u4e00\u4e2a\u5de5\u5177，\u6bd4\u5982 get_all_steps<\/p>\n<\/li>\n
\u53f3\u4fa7\u9762\u677f\u4f1a\u663e\u793a\u8be5\u5de5\u5177\u7684\u8be6\u7ec6\u4fe1\u606f\u3002\u56e0\u4e3a get_all_steps \u4e0d\u9700\u8981\u53c2\u6570，\u4f60\u53ea\u9700\u70b9\u51fb \u201cRun Tool\u201d \u6309\u94ae\u5373\u53ef\u3002<\/p>\n<\/li>\n<\/ul>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120306-688a09fa3a038.webp\")
<\/p>\n
\u70b9\u51fb\u201cRun Tool\u201d\u540e，\u4f60\u670d\u52a1\u5668\u8fd4\u56de\u7684 JSON \u54cd\u5e94\u4f1a\u663e\u793a\u5728\u53f3\u4fa7\u9762\u677f\u4e2d\u3002\u4f60\u53ef\u4ee5\u68c0\u67e5\u6570\u636e\u662f\u5426\u6b63\u786e\u3001\u683c\u5f0f\u662f\u5426\u7b26\u5408\u9884\u671f，\u5e76\u67e5\u770b\u662f\u5426\u6709\u9519\u8bef\u3002\u5bf9\u4e8e\u9700\u8981\u53c2\u6570\u7684\u5de5\u5177（\u5982 query_step_data），\u586b\u5199\u53c2\u6570\u540e\u70b9\u51fb\u201cRun\u201d\u5373\u53ef\u540c\u6837\u6d4b\u8bd5\u3002<\/p>\n
\u901a\u8fc7\u8fd9\u4e9b\u6b65\u9aa4，\u4f60\u53ef\u4ee5\u6d4b\u8bd5 MCP \u670d\u52a1\u5668\u7684\u5404\u4e2a\u90e8\u5206，\u786e\u4fdd\u5176\u7a33\u5b9a\u5e76\u51c6\u5907\u597d\u4e0e Claude Desktop \u96c6\u6210\u3002<\/p>\n<\/p>\n
mcp dev \u547d\u4ee4\u542f\u52a8\u7684\u662f\u7528\u4e8e\u8c03\u8bd5\u7684\u672c\u5730\u5f00\u53d1\u670d\u52a1\u5668\u3002\u786e\u8ba4 Resources\u3001Tools \u548c Prompts \u529f\u80fd\u6b63\u5e38\u540e，\u53ef\u4ee5\u505c\u6b62\u6d4b\u8bd5\u670d\u52a1\u5668\u3002<\/p>\n
\u65b9\u6cd5\u662f\u56de\u5230\u8fd0\u884c\u8be5\u8fdb\u7a0b\u7684\u7ec8\u7aef\u7a97\u53e3，\u6309 Ctrl+C\u3002<\/p>\n
…
\nConnected MCP client to backing server transport
\nCreated client\/server transports
\nSet up MCP proxy
\n^C <\/p>\n
\u670d\u52a1\u5668\u6d4b\u8bd5\u5b8c\u6210\u540e，\u4e0b\u4e00\u6b65\u662f\u5728 Claude Desktop \u4e2d\u5b89\u88c5\u5b83，\u7531 Claude Desktop \u7ba1\u7406\u670d\u52a1\u5668\u7684\u8fd0\u884c\u3002<\/p>\n<\/p>\n
\u4e3a\u4e86\u66f4\u5b9e\u7528\u548c\u76f4\u89c2\u5730\u4e0e MCP \u670d\u52a1\u5668\u4ea4\u4e92，\u6211\u4eec\u5c06\u4f7f\u7528 Claude Desktop \u4f5c\u4e3a MCP \u5ba2\u6237\u7aef\u3002\u4e0e\u4f20\u7edf\u547d\u4ee4\u884c\u5de5\u5177\u4e0d\u540c，MCP \u670d\u52a1\u5668\u9700\u8981\u517c\u5bb9 Model Context Protocol \u7684\u5ba2\u6237\u7aef，\u4e0d\u80fd\u76f4\u63a5\u8fd0\u884c\u547d\u4ee4\u3002Claude Desktop \u63d0\u4f9b\u4e86\u8fd9\u6837\u7684\u5ba2\u6237\u7aef\u754c\u9762，\u4f7f\u6211\u4eec\u53ef\u4ee5\u53d1\u9001 MCP \u670d\u52a1\u5668\u53ef\u5904\u7406\u7684\u81ea\u7136\u8bed\u8a00\u67e5\u8be2\u3002<\/p>\n
\u5b89\u88c5：\u4f60\u53ef\u4ee5\u5728 Meet Claude \\\\ Anthropic \u514d\u8d39\u4e0b\u8f7d Claude Desktop\u3002\u514d\u8d39\u7248\u672c\u5b8c\u5168\u8db3\u591f\u672c\u6587\u6d4b\u8bd5\u4f7f\u7528\u3002<\/p>\n<\/p>\n
\u5b89\u88c5 Claude Desktop \u540e，\u6211\u4eec\u6765\u6ce8\u518c MCP \u670d\u52a1\u5668：<\/p>\n
# Install the server for Claude Desktop
\nmcp install apple_watch_mcp.py –name "Apple Health Steps" <\/p>\n
\u786e\u8ba4\u5b89\u88c5\u662f\u5426\u6210\u529f<\/p>\n
\u73b0\u5728，\u8ba9\u6211\u4eec\u786e\u8ba4\u547d\u4ee4\u662f\u5426\u6309\u9884\u671f\u6267\u884c\u3002mcp install \u547d\u4ee4\u4f1a\u5411\u4e00\u4e2a JSON \u914d\u7f6e\u6587\u4ef6\u6dfb\u52a0\u6761\u76ee\u3002\u5982\u679c\u4f60\u4f7f\u7528\u4e86 API \u5bc6\u94a5，\u914d\u7f6e\u4e2d\u4f1a\u5305\u542b\u76f8\u5e94\u7684\u73af\u5883\u53d8\u91cf\u3002\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u67e5\u770b\u8be5\u6587\u4ef6\u5185\u5bb9\u3002<\/p>\n
cat ~\/Library\/Application\\\\ Support\/Claude\/claude_desktop_config.json <\/p>\n
\u4f60\u5e94\u8be5\u4f1a\u5728\u6587\u4ef6\u4e2d\u770b\u5230\u7c7b\u4f3c\u5982\u4e0b\u7ed3\u6784，\u4f4d\u4e8e "mcpServers" \u952e\u4e0b：<\/p>\n
{
\n "mcpServers": {
\n "Apple Health Steps": {
\n "command": "\/full\/path\/to\/your\/uv",
\n "args": [
\n "–directory",
\n "\/path\/to\/repository\/apple-watch-health-mcp",
\n "run",
\n "apple_watch_mcp.py"
\n \/\/ Other arguments may appear depending on your setup
\n ],
\n "env": {
\n "ES_API_KEY": "YOUR_ES_API_KEY"
\n }
\n }
\n \/\/ … other servers might be listed here
\n }
\n} <\/p>\n
\u5982\u679c\u4f60\u627e\u5230\u4e86 \u201cApple Health Steps\u201d \u8fd9\u4e00\u5757，\u606d\u559c！\u4f60\u7684\u670d\u52a1\u5668\u5df2\u6210\u529f\u914d\u7f6e\u3002<\/p>\n
\u91cd\u542f\u63d0\u9192<\/p>\n
\u91cd\u8981：\u5b89\u88c5\u670d\u52a1\u5668\u540e，\u9700\u8981\u91cd\u542f Claude Desktop \u4ee5\u52a0\u8f7d\u65b0\u914d\u7f6e\u3002<\/p>\n
\u53ea\u6709\u6253\u5f00 Claude \u65f6\u670d\u52a1\u5668\u624d\u4f1a\u542f\u52a8\u3002<\/p>\n<\/li>\n
\u5982\u679c\u51fa\u73b0\u6267\u884c\u9519\u8bef，\u53ef\u4ee5\u76f4\u63a5\u5728 Claude Desktop \u67e5\u770b\u65e5\u5fd7\u3002<\/p>\n<\/li>\n<\/ul>\n
\u914d\u7f6e\u5b8c\u6210\u540e，\u6700\u6709\u8da3\u7684\u90e8\u5206\u6765\u4e86：\u4f7f\u7528\u81ea\u7136\u8bed\u8a00\u4e0e\u4f60\u7684 MCP \u670d\u52a1\u5668\u5bf9\u8bdd\u3002<\/p>\n<\/p>\n
\u8981\u4e0e\u65b0\u670d\u52a1\u5668\u4ea4\u4e92，\u9700\u8981\u8fdb\u5165\u5ba2\u6237\u7aef\u754c\u9762\u3002\u6253\u5f00 Claude Desktop \u5e94\u7528，\u786e\u4fdd\u4f60\u5904\u4e8e\u4e3b\u804a\u5929\u754c\u9762\u3002<\/p>\n
\u8fd9\u91cc\u662f\u4f60\u63d0\u95ee\u7684\u5730\u65b9\u3002\u5c4f\u5e55\u5e94\u5982\u4e0b\u56fe\u6240\u793a，\u6587\u672c\u6846\u5185\u5df2\u7ecf\u6709\u6211\u4eec\u7684\u7b2c\u4e00\u4e2a\u95ee\u9898\u51c6\u5907\u597d\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120306-688a09fa47f8d.webp\")
<\/p>\n
\u4e00\u5207\u51c6\u5907\u5c31\u7eea，\u5f00\u59cb\u5427\u3002<\/p>\n<\/p>\n
\u73b0\u5728，\u53ea\u9700\u6309 \u201cEnter\u201d \u53d1\u9001\u4f60\u8f93\u5165\u7684\u95ee\u9898\u3002\u8fd9\u662f RAG \u7684\u4e00\u4e2a\u5f88\u597d\u793a\u4f8b，\u6a21\u578b\u4f1a\u68c0\u7d22\u7279\u5b9a\u6570\u636e\u8fdb\u884c\u56de\u7b54\u3002<\/p>\n
\u4f60\u8f93\u5165：<\/p>\n
How many steps did I take last Thursday? <\/p>\n
\u671f\u671b\u7ed3\u679c：<\/p>\n
Last Thursday (June 5th), you took 5,360 steps total. Your step count ranged from 450 to 2,340 steps across 4 different recordings throughout the day.<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120306-688a09fa6adf3.webp\")
<\/h4>\n\u8fd9\u5f88\u6709\u7528，\u4f46 MCP \u7684\u771f\u6b63\u529b\u91cf\u5728\u4e8e\u5b83\u4e0d\u4ec5\u80fd\u8fd4\u56de\u6587\u672c\u3002<\/p>\n<\/p>\n
\u73b0\u5728，\u8ba9\u6211\u4eec\u8bf7\u6c42\u670d\u52a1\u5668\u5904\u7406\u6570\u636e\u5e76\u4ee5\u7ed3\u6784\u5316\u683c\u5f0f\u8fd4\u56de，\u6bd4\u5982\u67f1\u72b6\u56fe\u3002<\/p>\n
\u4f60\u8f93\u5165：<\/p>\n
Show me a summary of my step count from the last 7 days as a bar chart, including some insights on it. <\/p>\n
\u56e0\u4e3a\u4f60\u7684\u670d\u52a1\u5668\u7406\u89e3\u610f\u56fe\u5e76\u6709\u5de5\u5177\u6765\u805a\u5408\u6570\u636e，\u5b83\u4e0d\u4ec5\u8fd4\u56de\u4e00\u53e5\u8bdd\u3002\u5b83\u751f\u6210\u7ed3\u6784\u5316\u6570\u636e，\u4ee5\u4fbf\u5ba2\u6237\u7aef（ Claude ）\u53ef\u4ee5\u6e32\u67d3\u53ef\u89c6\u5316\u3002<\/p>\n
![\"\"](\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2025\/07\/20250730120306-688a09fa95ef8.webp\")
<\/p>\n
\u606d\u559c！<\/p>\n
\u4f60\u5df2\u6210\u529f\u521b\u5efa\u4e00\u4e2a\u529f\u80fd\u6027\u7684 MCP \u670d\u52a1\u5668，\u5c06 Claude \u8fde\u63a5\u5230\u4f60\u7684\u5065\u5eb7\u6570\u636e，\u4f7f\u7528 Elasticsearch \u4f5c\u4e3a\u540e\u7aef，\u5e76\u652f\u6301\u81ea\u7136\u8bed\u8a00\u4ea4\u4e92\u3002<\/p>\n<\/p>\n
\u5728\u8fd9\u7bc7\u535a\u5ba2\u4e2d，\u6211\u4eec\u63a2\u8ba8\u4e86\u5982\u4f55\u521b\u5efa\u4e00\u4e2a\u5c06 Apple HealthKit \u6570\u636e\u4e0e LLM \u4ee3\u7406\u96c6\u6210\u7684 MCP \u670d\u52a1\u5668，\u4f7f\u7528 FastMCP \u548c Elasticsearch\u3002\u6211\u4eec\u8fd8\u5c55\u793a\u4e86\u5982\u4f55\u4f7f\u7528 MCP Inspector \u548c Claude Desktop \u672c\u5730\u6d4b\u8bd5\u670d\u52a1\u5668\u3002\u4ee3\u7801\u5728\u8fd9\u91cc\u53ef\u7528\u3002<\/p>\n
\u8fd9\u79cd\u65b9\u6cd5\u5141\u8bb8 AI \u52a9\u624b\u4ee5\u4e0a\u4e0b\u6587\u76f8\u5173\u3001\u5b89\u5168\u4e14\u76f4\u89c2\u7684\u65b9\u5f0f\u8bbf\u95ee\u548c\u5206\u6790\u5065\u5eb7\u6570\u636e\u3002<\/p>\n<\/p>\n
\u9644\u52a0\u8d44\u6e90：<\/p>\n
\u539f\u6587：Building an MCP server with Elasticsearch for real health data – Elasticsearch Labs<\/p>\n","protected":false},"excerpt":{"rendered":"
\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e12\u6b21\uff0c\u6536\u85cf8\u6b21\u3002\u6458\u8981\uff1a\u672c\u6587\u4ecb\u7ecd\u4e86\u5982\u4f55\u4f7f\u7528FastMCP\u6846\u67b6\u548cElasticsearch\u6784\u5efa\u4e00\u4e2aMCP\uff08Model Context Protocol\uff09\u670d\u52a1\u5668\u6765\u7ba1\u7406\u548c\u5206\u6790Apple Health\u5065\u5eb7\u6570\u636e\u3002\u6587\u7ae0\u8be6\u7ec6\u8bb2\u89e3\u4e86MCP\u7684\u4e09\u4e2a\u6838\u5fc3\u7ec4\u4ef6\uff08Resources\u3001Tools\u548cPrompts\uff09\u7684\u5b9e\u73b0\u65b9\u6cd5\uff0c\u5c55\u793a\u4e86\u5982\u4f55\u5c06Elasticsearch\u4f5c\u4e3a\u6570\u636e\u5b58\u50a8\u540e\u7aef\uff0c\u5e76\u901a\u8fc7Claude Desktop\u5b9e\u73b0\u81ea\u7136\u8bed\u8a00\u4ea4\u4e92\u67e5\u8be2\u3002\u8be5\u89e3\u51b3\u65b9\u6848\u652f\u6301\u52a8\u6001\u6570\u636e\u67e5\u8be2\u3001\u8d8b\u52bf\u5206\u6790\u548c\u53ef\u89c6\u5316\u5c55\u793a\uff0c\u4e3aLLM\u4ee3\u7406\u63d0\u4f9b\u4e86\u5b9e\u65f6\u5065\u5eb7\u6570\u636e\u8bbf\u95ee\u80fd\u529b\u3002\u6587\u4e2d\u5305\u542b\u5b8c\u6574\u7684\u4ee3\u7801\u5b9e\u73b0\u3001\u6d4b\u8bd5\u65b9_elasticsearch mcp<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[66,4416,1299,73,50,3368,323,2215],"topic":[],"class_list":["post-49832","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-elastic","tag-elasticsearch","tag-mcp","tag-50","tag-3368","tag-323","tag-2215"],"yoast_head":"\n