kb-mcp-server

kb-mcp-server

Enables semantic search over a local knowledge base using MCP tools, allowing AI clients to retrieve relevant document chunks via the search_knowledge tool.

Category
Visit Server

README

kb-mcp-server

自托管知识库的 TypeScript/Node monorepo。文档经分块与向量化后存入本地 Chroma;AI 客户端通过 MCP(stdio 或 Streamable HTTP)进行语义检索,文档的上传与管理走 REST API / CLI(Web 管理界面在 Phase 4 规划中)。

功能概览

能力 状态
文档入库(txt / markdown / 文本层 PDF)
CherryIn 嵌入(qwen/qwen3-embedding-8b
本地 Chroma 向量存储
REST API(上传、列表、搜索、健康检查)
MCP 工具 search_knowledge(stdio + HTTP)
Web 管理界面、可选 API Key 鉴权 🔜 Phase 4

架构

┌─────────────┐     ingest      ┌──────────────┐     embed      ┌─────────┐
│ CLI / REST  │ ──────────────► │ @kb/core     │ ─────────────► │ Chroma  │
│  (backend)  │                 │ Ingestion    │                │ :8000   │
└─────────────┘                 └──────────────┘                └────▲────┘
                                                                   │
┌─────────────┐     search      ┌──────────────┐                   │
│ MCP Client  │ ──────────────► │ SearchService│ ──────────────────┘
│ (stdio/HTTP)│                 │ (@kb/core)   │
└─────────────┘                 └──────────────┘
端口 服务 说明
8000 Chroma 向量数据库(HTTP sidecar)
3000 Backend Fastify REST API + Swagger UI
3100 MCP HTTP Streamable HTTP,POST /mcp
MCP stdio Cursor 等本地客户端,无端口

MCP 层仅负责检索;入库、删除等管理操作通过 Backend 或 CLI 完成。

环境要求

  • Node.js 24.x
  • pnpm 11.x
  • Python 3(Windows 上运行 Chroma:pnpm setup:chroma 安装 chromadb
  • CherryIn API Keyopen.cherryin.cc

快速开始

1. 安装依赖

git clone https://github.com/laneinrain/kb-mcp-server.git
cd kb-mcp-server
pnpm install

2. 配置环境变量

cp .env.example .env
# 编辑 .env,至少填写 CHERRYIN_API_KEY

3. 安装 Chroma(Windows 推荐 Python 方式)

pnpm setup:chroma

4. 构建并启动开发环境

pnpm build
pnpm dev

pnpm dev 会依次启动 Chroma、Backend(:3000)和 MCP HTTP(:3100)。

5. 入库文档

pnpm ingest path/to/document.pdf
pnpm ingest path/to/notes.md --collection default

或通过 REST 上传:

curl -F "file=@./scripts/fixtures/sample.md" http://127.0.0.1:3000/api/v1/documents

6. 验证搜索

REST:

curl -X POST http://127.0.0.1:3000/api/v1/search \
  -H "Content-Type: application/json" \
  -d '{"query": "your question", "topK": 5}'

健康检查:

curl http://127.0.0.1:3000/health
curl http://127.0.0.1:3000/health/chroma
curl http://127.0.0.1:3000/health/embeddings

API 文档:http://127.0.0.1:3000/docs

MCP 配置

MCP 暴露单一工具 search_knowledge

参数 类型 说明
query string 搜索问题(必填,1–2000 字符)
top_k number 返回条数,1–10,默认 5
collection string 集合名,默认 default

方式一:stdio(Cursor 本地)

先构建 MCP 包:

pnpm --filter @kb/mcp-server build

在 Cursor 的 MCP 配置(项目 .cursor/mcp.json 或全局 ~/.cursor/mcp.json)中添加:

{
  "mcpServers": {
    "kb-mcp-server": {
      "command": "C:/Program Files/nodejs/node.exe",
      "args": ["D:/project_ai/kb-mcp-server/apps/mcp-server/dist/stdio.js"],
      "cwd": "D:/project_ai/kb-mcp-server"
    }
  }
}

Windows 提示

  • 将路径改为你本机的项目绝对路径。
  • Cursor GUI 进程可能找不到 node,请使用 node.exe 的完整路径。
  • 必须设置 cwd,以便正确加载项目根目录的 .env
  • 也可使用 pnpm --filter @kb/mcp-server dev:stdio,但需保证 cwd 为 monorepo 根目录。

方式二:Streamable HTTP(远程 / 多客户端)

先运行 pnpm dev 或单独启动 MCP HTTP:

pnpm --filter @kb/mcp-server dev

Cursor MCP 配置示例:

{
  "mcpServers": {
    "kb-mcp-server": {
      "url": "http://127.0.0.1:3100/mcp"
    }
  }
}

浏览器直接访问 GET /mcp 会返回 404,这是预期行为;MCP 仅接受 POST /mcp

环境变量

完整示例见 .env.example。常用项:

变量 默认值 说明
CHERRYIN_API_KEY CherryIn 嵌入 API 密钥(必填)
CHROMA_HOST / CHROMA_PORT localhost / 8000 Chroma 地址
BACKEND_HOST / BACKEND_PORT 127.0.0.1 / 3000 REST API
MCP_HTTP_HOST / MCP_HTTP_PORT 127.0.0.1 / 3100 MCP HTTP
DEFAULT_COLLECTION default 默认向量集合
EMBEDDING_MODEL qwen/qwen3-embedding-8b 嵌入模型
CHUNK_SIZE / CHUNK_OVERLAP 1024 / 154 分块参数
DATA_DIR ./data 本地数据目录(SQLite、上传缓存)

切勿将 .env 提交到 Git。

REST API 摘要

方法 路径 说明
GET /health 服务存活
GET /health/chroma Chroma 连通性
GET /health/embeddings 嵌入 API 连通性
POST /api/v1/documents 上传文档(multipart)
GET /api/v1/documents 列出已入库文档
GET /api/v1/documents/:id 文档详情
DELETE /api/v1/documents/:id 删除文档及向量
POST /api/v1/search 语义搜索

项目结构

kb-mcp-server/
├── apps/
│   ├── backend/          # Fastify REST API
│   └── mcp-server/       # MCP stdio + Streamable HTTP
├── packages/
│   ├── config/           # 环境变量与常量
│   └── core/             # 入库、检索、Chroma、嵌入客户端
├── scripts/
│   ├── ingest.ts         # CLI 入库
│   ├── start-chroma.ts   # 启动 Chroma sidecar
│   └── wait-for-chroma.ts
├── .env.example
└── package.json          # pnpm workspace 根

开发与测试

# 全量构建
pnpm build

# 全量测试
pnpm test

# 单独启动各服务
pnpm --filter @kb/backend dev
pnpm --filter @kb/mcp-server dev          # HTTP :3100
pnpm --filter @kb/mcp-server dev:stdio    # stdio

# 等待 Chroma 就绪
pnpm wait:chroma

路线图

  • Phase 1–3(已完成):平台基础、REST 检索、MCP 检索服务
  • Phase 4(规划中):Web 管理界面、可选 API Key 鉴权、CLI 增强

许可证

尚未指定开源许可证。如需公开协作,请自行添加 LICENSE 文件。

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured