AI agents on Polymarket data — MCP, OpenAPI, and llms.txt

Three integration paths

All three use the same rm_ API key. Pick whichever matches the runtime you are wiring up.

1. Native MCP (Claude Desktop, Claude Code, Cursor, Windsurf, Cline)

The Resolved Markets MCP server exposes six tools — list_markets, get_orderbook, get_snapshot, query_snapshots, get_market_summary, get_system_stats — and two resources (markets://live, prices://latest). Configure once, and any MCP-capable client can call them as first-class tools with no glue code.

# Claude Desktop / Claude Code config (claude_desktop_config.json)
{
  "mcpServers": {
    "resolved-markets": {
      "command": "npx",
      "args": ["-y", "@elcara-hq/resolvedmarkets-mcp"],
      "env": {
        "HF_API_URL": "https://api.resolvedmarkets.com",
        "HF_API_KEY": "rm_your_key_here"
      }
    }
  }
}

HTTP transport is also supported (MCP_TRANSPORT=http MCP_PORT=3002) for self-hosted agents that prefer remote MCP over stdio.

2. Tool-use / function-calling with the OpenAPI spec

The OpenAPI 3.1 spec at https://resolvedmarkets.com/openapi.json can be imported into Postman, Insomnia, RapidAPI, GPT Actions, or any code generator. For Anthropic and OpenAI tool-use, expose the operations in the spec as your tool definitions and the model will produce typed calls automatically.

# Python — load the spec into an OpenAI function-calling tool
import requests, json

spec = requests.get("https://resolvedmarkets.com/openapi.json").json()
tools = []
for path, ops in spec["paths"].items():
    for method, op in ops.items():
        tools.append({
            "type": "function",
            "function": {
                "name": op["operationId"],
                "description": op.get("summary", ""),
                "parameters": op.get("requestBody", {}).get("content", {})
                  .get("application/json", {}).get("schema", {}),
            },
        })

3. Plain context — paste llms.txt into the system prompt

For chat assistants without tool-use (or for one-shot questions), paste /llms.txt (~10 KB summary) or /llms-full.txt (~43 KB full reference) into the system prompt. The model can then answer questions about endpoints, parameters, and response shapes — and produce correct curl or Python that hits real URLs.

Drop-in system prompt

Copy this into any LLM's system prompt to give it grounded knowledge of the Resolved Markets API. The agent will quote the right endpoints and stop hallucinating routes.

You have access to live and historical Polymarket prediction market and Hyperliquid perpetual futures orderbook data via the Resolved Markets API.

API base URL: https://api.resolvedmarkets.com
Authentication: header "X-API-Key: rm_..." (the user will provide their key)

Reference docs (read these first if you need details):
- Short reference: https://resolvedmarkets.com/llms.txt
- Full reference: https://resolvedmarkets.com/llms-full.txt
- OpenAPI spec: https://resolvedmarkets.com/openapi.json

Coverage: crypto (BTC, ETH, SOL, XRP), sports (NBA, NFL, EPL), economics (FOMC, NFP), weather (~30 city / climate markets), and Hyperliquid perpetual futures. Snapshots include full bid/ask depth, mid, spread, sequence numbers, and the paired crypto spot price at capture time.

When answering questions about Polymarket markets, prefer calling these endpoints over guessing:
- GET /v1/markets/live           — list active markets (filter with ?category= and ?subcategory=)
- GET /v1/markets/:id/orderbook  — live UP and DOWN orderbook for a market
- GET /v1/markets/:id/snapshots  — historical snapshots (?from= ?to= ?limit=500)
- GET /v1/markets/:id/summary    — aggregated stats: spread, depth, price range
- GET /v1/public-stats           — platform-wide stats (no auth required)

Probabilities are USDC prices in 0.01..0.99. Each market has UP and DOWN tokens; UP + DOWN ≈ 1.00.

Compatible runtimes

• MCP-native clients — Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Continue, Zed.
• Anthropic SDK — tool-use with the OpenAPI operations as tool definitions.
• OpenAI SDK — function calling, GPT Actions for ChatGPT custom GPTs.
• Google AI SDK — Gemini function calling.
• Frameworks — LangChain, LlamaIndex, CrewAI, AutoGen, Haystack, Semantic Kernel.
• Plain HTTP — anything that can send X-API-Key and parse JSON.

60-second quickstart (Python)

import os, requests

BASE = "https://api.resolvedmarkets.com"
H = {"X-API-Key": os.environ["RM_API_KEY"]}

# 1. Find an active BTC market
markets = requests.get(f"{BASE}/v1/markets/live",
                       params={"category": "crypto", "subcategory": "BTC"},
                       headers=H).json()
market_id = markets["markets"][0]["condition_id"]

# 2. Pull live orderbook
ob = requests.get(f"{BASE}/v1/markets/{market_id}/orderbook", headers=H).json()
print(ob["up"]["best_bid"], ob["up"]["best_ask"])

# 3. Pull last hour of snapshots
snaps = requests.get(f"{BASE}/v1/markets/{market_id}/snapshots",
                     params={"limit": 500}, headers=H).json()