In the world of AI agents, accessing live data from prediction markets like Polymarket has been a challenge. While Polymarket offers a public API, AI tools such as Claude Desktop, Cursor, or Cline don't natively speak it. That's where the polymarket-mcp-pro server steps in—a free, open-source Model Context Protocol (MCP) server that bridges the gap. Below, we answer common questions about this tool, its capabilities, and how it empowers your AI agent with real-time market data.

What exactly is the Polymarket MCP server, and why was it built?

The Polymarket MCP server, named polymarket-mcp-pro, is a lightweight interface that translates an AI agent's requests into Polymarket API calls. Before its creation, asking an agent about prediction markets often resulted in hallucinations, stale data from outdated sources, or polite refusals. The server solves this by exposing seven read-only tools that any MCP-compatible client can use. It was built out of frustration—the developer wanted a reliable way for agents to fetch real market data without embedding HTTP code in prompts. The result is a structured JSON tool surface that lets agents query markets, prices, crashes, and more, all while remaining completely read-only. No orders, no wallets—just observation.

What data powers the MCP server?

Behind the scenes, the server relies on the api.protodex.io data layer, which indexes a massive dataset. As of now, it includes over 13,964 Polymarket markets, more than 10.8 million price snapshots taken every 15 minutes per active market, and cumulative volume exceeding $11.7 billion. This data is continuously refreshed and served through a stable REST API. When you ask your agent to identify markets that dropped 15% in the last four hours, it queries this entire corpus—not a tiny sample. This depth ensures your agent delivers accurate, actionable insights based on real, comprehensive data.

How do I install the polymorphic MCP server?

Installation is quick, with two primary methods. If you have uv installed, run the command: uvx --from polymarket-mcp-pro polymarket-mcp. Alternatively, using standard pip: pip install polymarket-mcp-pro. Note the PyPI distribution name is polymarket-mcp-pro because the bare name was taken. The CLI and import path remain polymarket_mcp. That's it—the server is ready to connect to your AI agent.

How do I configure the server with Claude Desktop or Cursor?

To use it with Claude Desktop, edit the configuration file. On macOS, it's at ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows, at %APPDATA%\Claude\claude_desktop_config.json. Add the following JSON block under mcpServers:

{
  "mcpServers": {
    "polymarket": {
      "command": "uvx",
      "args": ["polymarket-mcp"]
    }
  }
}

Restart Claude Desktop, and the Polymarket tools appear in the tool picker. For Cursor, use the same JSON in Settings → Features → MCP Servers. Other MCP clients like Cline or Continue work similarly using the same stdio command.

What tools does the server provide?

The server exposes seven tools, all read-only. get_stats returns dataset-wide counts and freshness. list_markets gives a paginated list with category, search, and sort options. get_market provides detailed info on a single market, including CLOB token IDs. get_prices fetches historical price snapshots for Yes/No outcomes. get_crashes identifies markets that dropped at least N% in the last H hours. get_categories shows category breakdowns with market counts and volume. Finally, get_orderbook delivers live bid/ask depth for a given CLOB token ID. All tools are strictly observational—no order placement or wallet interaction.

How can I use the crash scanner feature?

One of the most practical tools is get_crashes. You can set a threshold and time window—for example, scanning for markets that fell 15% within the last 4 hours and filtering by volume greater than $100,000. The agent returns a structured list of the top crashes, along with a brief summary of each market. Many days you'll see zero matches, but occasionally something interesting pops up. It's excellent for a morning routine or monitoring breaking news. For deeper analysis, combine with get_market or get_orderbook to explore specific opportunities.

What real-world workflows does the server enable?

Beyond crash scanning, the server excels at researching news events. For instance, when a major story breaks, you can ask the agent to list markets with high volume and then pull historical prices to see how odds changed. This helps you understand market sentiment quickly. Another workflow is periodic monitoring: set the agent to check certain markets at intervals and summarize shifts. Because the tools are read-only, you can safely run them without risking trades. The developer uses it daily for quick checks, focusing on markets with significant movement or high liquidity. It transforms your AI agent from a guessing machine into a data-driven analyst.