TTG Scratchpad MCP Server
Provides persistent agent workspaces with MongoDB storage for file management and real-time task progress tracking. It enables secure, isolated user environments with full activity logging for workspace operations.
README
TTG Scratchpad MCP Server
Production-ready MCP server for agent workspace/scratchpad functionality with MongoDB persistence and user isolation.
Features
- Persistent Workspaces: Tasks and files stored in MongoDB Atlas
- User Isolation: Each user has their own isolated workspace (via X-User-ID header)
- File Management: Create, read, update, delete files in workspace
- Progress Tracking: Real-time progress updates visible in chat widget
- Activity Logging: Full audit trail of workspace operations
- Bearer Authentication: Secure API access via MCP_API_KEYS
Tools
Workspace Management
scratchpad_start- Start a new task workspacescratchpad_update- Update progress during taskscratchpad_complete- Mark task as completescratchpad_demo- Demo widget with sample data
File Management
scratchpad_write_file- Save file to workspacescratchpad_read_file- Read file from workspacescratchpad_list_files- List all workspace filesscratchpad_delete_file- Delete file from workspace
Local Development
Prerequisites
- Python 3.11+
- MongoDB Atlas cluster (ttg-workspaces)
Setup
# Create virtual environment
python -m venv venv
.\venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Copy and configure environment
copy .env.template .env
# Edit .env with your MongoDB URI and API key
Run Locally
cd C:\Users\Adam\Desktop\ttgeco\scratchpad-mcp
.\venv\Scripts\activate
python server.py
Server runs on http://localhost:8001/mcp
Test Without Database
The server runs in demo/stateless mode if MONGODB_URI is not set. Useful for UI testing.
Deployment (Render)
1. Push to GitHub
git add .
git commit -m "Add MongoDB persistence and file management"
git push origin main
2. Configure Render Environment Variables
In Render dashboard, add these environment variables:
| Variable | Value | Notes |
|---|---|---|
MONGODB_URI |
mongodb+srv://... |
Your ttg-workspaces connection string |
MCP_API_KEYS |
scratchpad-mcp-ttg-2026-... |
Must match LibreChat SCRATCHPAD_MCP_TOKEN |
Do NOT set PORT - Render provides it automatically.
3. Verify Deployment
After deploy, test the /mcp endpoint:
curl -X POST https://scratchpad-mcp-18ab.onrender.com/mcp \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_TOKEN"
LibreChat Configuration
librechat.yaml
mcpServers:
scratchpad:
type: streamable-http
url: "https://scratchpad-mcp-18ab.onrender.com/mcp"
headers:
Authorization: "Bearer ${SCRATCHPAD_MCP_TOKEN}"
X-User-ID: "{{LIBRECHAT_USER_ID}}"
timeout: 60000
initTimeout: 30000
serverInstructions: true
LibreChat .env
SCRATCHPAD_MCP_TOKEN=your-token-here
Database Schema
Collections
workspaces
{
"_id": "ObjectId",
"user_id": "string",
"conversation_id": "string (optional)",
"task": "string",
"status": "idle|active|complete",
"progress": {"current": 0, "total": 10},
"working_on": "string",
"created_at": "datetime",
"updated_at": "datetime"
}
files
{
"_id": "ObjectId",
"user_id": "string",
"workspace_id": "ObjectId",
"name": "string",
"path": "string",
"type": "file|folder",
"content": "string",
"updated": true,
"created_at": "datetime",
"updated_at": "datetime"
}
activity_logs
{
"_id": "ObjectId",
"user_id": "string",
"workspace_id": "ObjectId",
"action": "string",
"icon": "write|read|complete|delete",
"timestamp": "datetime"
}
Security
- Bearer token authentication via
MCP_API_KEYS - User isolation via
X-User-IDheader (provided by LibreChat) - All database queries scoped by user_id
- Input validation on file paths and content
Troubleshooting
"No HTTP request context available"
Tool called outside of request context. Ensure using FastMCP 0.5.0+.
"X-User-ID header required"
LibreChat not sending user ID. Check librechat.yaml headers config.
"MONGODB_URI environment variable required"
Set MONGODB_URI in Render dashboard or local .env.
Connection timeout
Check MongoDB Atlas network access - ensure Render IP is whitelisted or use 0.0.0.0/0 for all access.
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。