Meeting Notes MCP Server
A self-hosted MCP server for storing, searching, and querying meeting notes with semantic search and RAG capabilities using SQLite + OpenRouter.
README
Meeting Notes MCP Server
A self-hosted Model Context Protocol (MCP) server for storing, searching, and querying meeting notes with semantic search and RAG (Retrieval-Augmented Generation) capabilities using SQLite + OpenRouter.
Description
This MCP server provides a complete meeting notes management system with the following capabilities:
- Store meeting notes with title, date, participants, tags, and full content
- Semantic search using vector embeddings (sqlite-vec) to find meetings by meaning, not just keywords
- RAG-powered Q&A — Ask natural-language questions about your meetings and get contextually accurate answers
- AI summaries — Generate brief, detailed, or bullet-point summaries of any meeting
- Flexible filtering — Filter meetings by date range, participant, or tag
- Persistent storage — All data is stored in SQLite with vector embeddings
Built With
| Component | Technology |
|---|---|
| Runtime | Node.js + TypeScript |
| Database | SQLite + sqlite-vec (vector extension) |
| MCP Framework | @modelcontextprotocol/sdk |
| AI API | OpenRouter (embeddings + chat completions) |
| Validation | Zod |
| Testing | Jest + ts-jest |
Installation
Prerequisites
- Node.js >= 18.0.0
- npm or yarn
Steps
-
Clone the repository
git clone <repository-url> cd meeting-notes-mcp -
Install dependencies
npm install -
Create configuration file
cp .env.example .env -
Configure your environment variables
Edit
.envand add your OpenRouter API key:OPENROUTER_API_KEY=sk-or-v1-your-key-hereOptional overrides:
DATABASE_PATH=./data/meetings.db EMBEDDING_MODEL=openai/text-embedding-3-small CHAT_MODEL=anthropic/claude-sonnet-4-20250514 MAX_CONTEXT_MEETINGS=5Optional authentication (HTTP mode only):
MCP_USERNAME=admin MCP_PASSWORD=changeme JWT_SECRET=your-super-secret-jwt-key-min-32-chars-longNote: If any of the three auth variables is missing, authentication is disabled and the server behaves exactly as before. The stdio transport is never affected.
-
Build the project
npm run build
Docker Deployment (Recommended for VPS)
Deploy the server as a Docker container — ideal for VPS or remote servers.
Prerequisites
- Docker & Docker Compose installed on your VPS
Build & Run with Docker
-
Build the image
docker build -t meeting-notes-mcp . -
Run the container
docker run -d \ --name meeting-notes-mcp \ -p 3000:3000 \ -e OPENROUTER_API_KEY=sk-or-v1-your-key-here \ -e MCP_HTTP_PORT=3000 \ -e MCP_USERNAME=admin \ -e MCP_PASSWORD=changeme \ -e JWT_SECRET=your-super-secret-jwt-key-min-32-chars-long \ -v meeting-notes-data:/data \ meeting-notes-mcp-p 3000:3000— Exposes the MCP HTTP server on port 3000-e OPENROUTER_API_KEY=...— Required API key-e MCP_HTTP_PORT=3000— Enables HTTP mode (without this, it runs stdio)-e MCP_USERNAME,-e MCP_PASSWORD,-e JWT_SECRET— Optional auth credentials (omit to disable)-v meeting-notes-data:/data— Persists the SQLite database in a Docker volume
-
Verify it's running
docker logs meeting-notes-mcpYou should see:
Meeting Notes MCP HTTP Server listening on http://0.0.0.0:3000/mcp
Docker Compose ( easiest )
Create a docker-compose.yml:
services:
meeting-notes:
build: .
container_name: meeting-notes-mcp
ports:
- "3000:3000"
environment:
- OPENROUTER_API_KEY=sk-or-v1-your-key-here
- MCP_HTTP_PORT=3000
- DATABASE_PATH=/data/meetings.db
- EMBEDDING_MODEL=openai/text-embedding-3-small
- CHAT_MODEL=anthropic/claude-sonnet-4-20250514
# Optional authentication (uncomment to enable)
# - MCP_USERNAME=admin
# - MCP_PASSWORD=changeme
# - JWT_SECRET=your-super-secret-jwt-key-min-32-chars-long
volumes:
- meeting-notes-data:/data
restart: unless-stopped
volumes:
meeting-notes-data:
Then run:
docker compose up -d
Usage
Running the Server
Stdio Mode (default, for local MCP clients)
npm start
For development with auto-rebuild:
npm run dev
MCP Client Configuration
Local Stdio Connection
Add the server to your MCP client configuration (e.g., Claude Desktop, Cursor, OpenCode):
{
"mcpServers": {
"meeting-notes": {
"type": "local",
"command": ["node", "/your/path/meeting_notes_mcp/dist/index.js"],
"env": {
"OPENROUTER_API_KEY": "sk-or-v1-your-key-here"
}
}
}
}
Remote HTTP Connection with Authentication (Optional)
If authentication is enabled on the server (MCP_USERNAME, MCP_PASSWORD, JWT_SECRET set), you must first obtain a token and then include it in all requests.
Step 1: Obtain a token
curl -X POST http://your-vps-ip:3000/auth/token \
-H "Content-Type: application/json" \
-d '{"username":"admin","password":"changeme"}'
Response:
{ "token": "eyJhbGciOiJIUzI1NiIs..." }
Step 2: Configure your MCP client with the token
{
"mcpServers": {
"meeting-notes": {
"type": "remote"
"url": "http://your-vps-ip:3000/mcp",
"headers": {
"Authorization": "Bearer eyJhbGciOiJIUzI1NiIs..."
}
}
}
}
Replace your-vps-ip with your actual VPS IP address or domain.
Security Note: For production, put the HTTP endpoint behind a reverse proxy (Nginx, Caddy, Traefik) with HTTPS. The built-in Bearer token authentication provides a lightweight protection layer for the MCP HTTP transport.
Available MCP Tools
The server exposes 6 tools for managing and querying meetings:
1. add_meeting
Store a new meeting note. Automatically generates an embedding for semantic search.
Example:
{
"title": "Product Roadmap Q3",
"date": "2025-07-15",
"participants": ["Alice", "Bob", "Charlie"],
"tags": ["product", "roadmap", "planning"],
"content": "Discussed Q3 priorities: launch search improvements, mobile app redesign, and AI-powered recommendations. Alice will own the search project, Bob will handle mobile, and Charlie will prototype the AI features."
}
2. get_meeting
Retrieve a meeting by its ID.
Example:
{ "id": 1 }
3. list_meetings
List meetings with optional filters.
Example:
{
"startDate": "2025-07-01",
"endDate": "2025-07-31",
"participant": "Alice",
"tag": "product"
}
4. search_meetings
Find meetings semantically using natural language. Returns the most contextually relevant meetings.
Example:
{
"query": "mobile app redesign plans",
"limit": 5
}
5. summarize_meeting
Generate an AI-powered summary of a meeting.
Example:
{
"id": 1,
"style": "bullets"
}
Style options: brief, detailed, bullets
6. ask_meetings
Ask a natural-language question about your meetings using RAG.
Example:
{
"question": "Who is responsible for the mobile app redesign?",
"maxMeetings": 5
}
The server embeds the question, finds similar meetings via vector search, and synthesizes an answer using the LLM.
Testing
Run the full test suite (189 tests):
npm test
Run with coverage:
npm test -- --coverage
Watch mode for development:
npm run test:watch
Project Structure
├── src/
│ ├── auth/ # JWT token creation and Express auth middleware
│ ├── config/ # Environment & configuration
│ ├── db/ # Database schema, connection, and CRUD operations
│ ├── services/ # OpenRouter client, embedding service
│ ├── tools/ # MCP tool implementations (6 tools)
│ ├── types/ # TypeScript interfaces and types
│ ├── validations/ # Zod schemas for input validation
│ ├── utils/ # Logging, retry logic, error handling, graceful degradation
│ ├── server.ts # Stdio MCP server startup
│ ├── http-server.ts # Streamable HTTP MCP server startup
│ └── __tests__/ # Jest test files
├── ai/
│ ├── SPECS.md # Project specification
│ ├── TASK.md # Development task tracker
│ └── SETUP.md # Client configuration guide
├── .env.example # Environment variable template
├── Dockerfile # Docker image for VPS deployment
├── docker-compose.yml # (Optional) Docker Compose setup
├── .gitignore
├── package.json
├── tsconfig.json
└── jest.config.js
Configuration Reference
| Variable | Default | Description |
|---|---|---|
OPENROUTER_API_KEY |
— | Your OpenRouter API key |
DATABASE_PATH |
./data/meetings.db |
Path to SQLite database |
EMBEDDING_MODEL |
openai/text-embedding-3-small |
Model for vector embeddings |
CHAT_MODEL |
anthropic/claude-sonnet-4-20250514 |
Model for summarization & RAG answers |
MAX_CONTEXT_MEETINGS |
5 |
Meetings retrieved for RAG context |
MCP_HTTP_PORT |
— | Set to enable HTTP mode (e.g., 3000) |
LOG_LEVEL |
info |
Logging verbosity: debug, info, warn, error |
MCP_USERNAME |
— | Single allowed username (optional, enables auth) |
MCP_PASSWORD |
— | Password for the single user (optional, enables auth) |
JWT_SECRET |
— | Secret key for signing/verifying JWTs (optional, enables auth) |
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。