PolyDesk
Enables automated research and trading on Polymarket prediction markets through a Notion dashboard, with AI-powered analysis and human-in-the-loop trade approval.
README
PolyDesk — Polymarket AI Research & Trading Control Plane
An MCP server that turns Polymarket prediction markets into a structured research and trading workflow — powered by Notion as the dashboard and knowledge base. Zero-interaction automation: check a box in Notion, AI research appears automatically.
Inspired by Karpathy's autoresearch: the AI agent runs iterative research loops on prediction markets, scores them, and writes structured findings to Notion for human review and trade execution.
Screenshots
Market Watchlist (left columns)

Market Watchlist (right columns) — Research checkbox triggers AI analysis

Research Reports Database

One-Command Setup via Claude Desktop

MCP Server Integration

Architecture
┌──────────────┐ ┌──────────────────┐ ┌──────────────┐
│ Polymarket │────→│ PolyDesk MCP │────→│ AI Agent │
│ Gamma API │ │ (this server) │ │ (Claude, etc)│
└──────────────┘ └──────────────────┘ └──────┬───────┘
│
▼
┌──────────────┐
│ Notion MCP │
│ (official) │
└──────┬───────┘
│
▼
┌──────────────┐
│ Notion │
│ Workspace │
│ (Dashboard) │
└──────────────┘
PolyDesk MCP provides Polymarket data, AI research (via local Ollama), and direct Notion integration. Notion MCP (official) provides additional Notion read/write capabilities. Together they create a zero-interaction trading control plane — all from Notion.
Key Features
- One-command setup —
call polydesk setupcreates the full Notion workspace with databases and trending markets - Zero-interaction research — Check the 🔬 Research box in Notion, AI analysis appears automatically (10s polling)
- Keyword search — Type a keyword (e.g. "Trump") in a new row, the system finds the matching Polymarket market
- Local AI — Research powered by Ollama (qwen2.5:14b) running locally, no cloud API needed
- Human-in-the-loop — AI generates signals, humans approve trades via Notion
Tools (22 total)
| Tool | Description |
|---|---|
setup |
Create Notion workspace with databases and trending markets |
scan_trending_markets |
Discover hottest markets by volume |
search_markets |
Search markets by keyword |
get_market |
Full details for a specific market |
get_events |
Browse top events (grouped markets) |
get_event |
Details for a specific event |
get_prices |
Bulk price feed for multiple markets |
auto_research_market |
Structured research prompt (autoresearch pattern) |
research_with_ollama |
Run local AI research on a market |
batch_research |
Research multiple markets in one loop |
calculate_trade |
Position sizing, R:R, Kelly criterion |
check_positions |
Live P&L for open positions |
compare_markets |
Side-by-side odds comparison |
edge_scanner |
Heuristic mispricing detector |
format_research_for_notion |
Generate rich Notion blocks from research |
format_watchlist_entry |
Format market data for Notion watchlist |
sync_notion_watchlist |
Sync watchlist with Polymarket prices |
watch |
Start background polling for Notion changes |
unwatch |
Stop background polling |
validate_human_overrides |
Check for human trade approvals |
generate_execution_plan |
Create trade execution plan |
Resources
| Resource | Description |
|---|---|
polydesk://schemas/notion-databases |
Notion database schemas for Watchlist and Research |
Prompts
| Prompt | Description |
|---|---|
setup-trading-desk |
Bootstrap the full Notion workspace |
daily-research-loop |
Run a complete research cycle on trending markets |
trade-review |
Sync positions and generate P&L summary |
How It Works
- Setup —
call polydesk setupcreates a Notion page with Watchlist and Research Reports databases, populated with live trending markets - Auto-watch — Background polling starts automatically, checking Notion every 10 seconds
- Research trigger — Check the 🔬 Research box on any market row → status changes to "Researching" → Ollama runs analysis → research report appears in the Research Reports database
- Keyword search — Add a new row with just a keyword → system searches Polymarket and fills in market data
- Human review — Review AI signals and set Human Approval to "Approved" → system generates execution plan
- Iterate — Re-check the research box to run deeper analysis iterations (ratchet mechanism)
Setup
Prerequisites
- Ollama with
qwen2.5:14bmodel (ollama pull qwen2.5:14b) - A Notion integration token
1. Install
npm install
npm run build
2. Configure Claude Desktop
Add both MCP servers to your claude_desktop_config.json:
{
"mcpServers": {
"polydesk": {
"command": "node",
"args": ["/path/to/polydesk-mcp/dist/index.js"],
"env": {
"NOTION_TOKEN": "ntn_YOUR_TOKEN"
}
},
"notion": {
"command": "npx",
"args": ["-y", "@notionhq/notion-mcp-server"],
"env": {
"OPENAPI_MCP_HEADERS": "{\"Authorization\": \"Bearer ntn_YOUR_TOKEN\", \"Notion-Version\": \"2022-06-28\"}"
}
}
}
}
3. Bootstrap Your Workspace
In Claude Desktop, say: "Call polydesk setup"
This creates the full Notion workspace and starts auto-watching.
4. Start Researching
Just check the 🔬 Research box on any market in Notion. No commands needed — research appears automatically.
You can also use Claude Desktop directly:
"Scan trending markets and research the top 5"
"Find markets about bitcoin and analyze the best opportunity"
"Run the daily research loop"
No Polymarket API Key Required
PolyDesk uses the public Polymarket Gamma API — no authentication needed for market data.
Development
npm run dev # Run with tsx (hot reload)
npm run check # Type-check
npm run build # Compile to dist/
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。