flowchart LR
    A[Claude Desktop<br/>Inspector<br/>Cursor]
    B[server.py<br/>FastMCP]
    C[kb.py<br/>纯函数]
    D[SiliconFlow<br/>BAAI/bge-m3<br/>1024 维]
    E[Supabase<br/>pgvector + RPC]

    A -- stdio JSON-RPC --> B
    B -- 调用 --> C
    C -- HTTPS embeddings --> D
    D -- 向量 --> C
    C -- RPC match_document_chunks<br/>+ documents 表 --> E
    E -- 检索结果 --> C
    C -- 结构化返回 --> B
    B -- JSON-RPC --> A

create extension if not exists vector;
create extension if not exists pgcrypto;

create table public.documents (
  id uuid primary key default gen_random_uuid(),
  user_id uuid not null references auth.users(id) on delete cascade,
  title text not null,
  content_type text not null check (content_type in ('pdf','txt','url','docx')),
  source_url text,
  status text not null default 'processing' check (status in ('processing','ready','failed')),
  error_message text,
  char_count int default 0,
  chunk_count int default 0,
  created_at timestamptz default now()
);
create index documents_user_id_idx on public.documents(user_id);

create table public.document_chunks (
  id uuid primary key default gen_random_uuid(),
  document_id uuid not null references public.documents(id) on delete cascade,
  user_id uuid not null references auth.users(id) on delete cascade,
  content text not null,
  embedding vector(1024),
  metadata jsonb default '{}'::jsonb,
  created_at timestamptz default now()
);
create index document_chunks_user_id_idx on public.document_chunks(user_id);
create index document_chunks_document_id_idx on public.document_chunks(document_id);
create index document_chunks_embedding_idx on public.document_chunks
  using ivfflat (embedding vector_cosine_ops) with (lists = 100);

create or replace function match_document_chunks(
  query_embedding vector(1024),
  tenant_id uuid,
  match_count int default 5,
  min_similarity float default 0.3
)
returns table (id uuid, document_id uuid, content text, similarity float, metadata jsonb)
language sql stable as $$
  select dc.id, dc.document_id, dc.content,
         1 - (dc.embedding <=> query_embedding) as similarity,
         dc.metadata
  from public.document_chunks dc
  where dc.user_id = tenant_id
    and 1 - (dc.embedding <=> query_embedding) > min_similarity
  order by dc.embedding <=> query_embedding
  limit match_count;
$$;

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

git clone https://github.com/<你的用户名>/kb-mcp-server.git
cd kb-mcp-server
Copy-Item .env.example .env

SUPABASE_URL=https://<your-project>.supabase.co
SUPABASE_SERVICE_ROLE_KEY=eyJ...
SILICONFLOW_API_KEY=sk-...
KB_TENANT_ID=<在你 Supabase auth.users 表里的 uuid>

npx @modelcontextprotocol/inspector uv run server.py

%APPDATA%\Claude\claude_desktop_config.json

uv run python client_demo.py

============================================================
✅ 已连接 kb-mcp-server(stdio 子进程)
============================================================

[1/4] list_tools → 工具清单:
  - search_knowledge_base: 语义检索知识库,返回与查询最相关的若干片段(含相似度与所属文档标题)。
  - list_documents: 列出当前知识库的全部文档及元数据(id、标题、类型、状态、片段数、创建时间)。
  - get_document_content: 按 document_id 拉取并拼接单篇文档全文。

[2/4] call_tool('search_knowledge_base', ...) →
  命中 3 条
  首条 similarity     = 0.7240
  首条 document_title = lantu-x10-manual
  首条 content 前 100 字:...

[3/4] read_resource('kb://documents') → 文档标题清单:
  - lantu-x10-manual(chunk_count=...)

[4/4] get_prompt('kb_qa', question='滤网怎么保养?') → 渲染后模板:
  [user] 基于知识库回答以下问题。请先调用 search_knowledge_base 检索相关片段...