Drop ParlayAPI into Claude Desktop, Cursor, or Claude Code in 30 seconds with our MCP server, or wire any other LLM to the REST API through the OpenAPI spec. Through MCP your agent gets 22 typed tools: live odds, player props, best line, consensus, bet verdicts, parlay grading, best-bets discovery, arbitrage, +EV and middles. CLV grading and SGP pricing are available over REST at /v1/clv and /v1/sgp/price.
Drop parlayapi-mcp into Cursor's settings. All 22 tools become callable from inside your editor.
pip install parlayapi-mcp5 minutes from pip install to "list +EV plays on tonight's MLB slate."
Connect to Claude Desktop via claude_desktop_config.json. Same MCP server, same 22 tools, available in every Claude Desktop conversation once configured.
{
"mcpServers": {
"parlay-api": {
"command": "parlayapi-mcp",
"env": {"PARLAY_API_KEY": "pak_live_..."}
}
}
}One-line install via the Claude Code CLI. Works global or scoped to a single project.
claude mcp add parlay-api \
--env PARLAY_API_KEY=pak_live_... \
-- parlayapi-mcpThe MCP server exposes 22 curated endpoints as typed tools, chosen so the whole set fits comfortably in a model's context. The agent's reasoning loop chains them automatically. Some prompts to try right away:
| Prompt | Tools chained |
|---|---|
| "What are the top 5 +EV plays on tonight's MLB slate?" | find_ev |
| "Are there any cross-book arbitrages on the NBA right now?" | arbitrage_scan |
| "Find totals middles open on NFL today, sort by window width" | middles_scan |
| "Score my bet Yankees -145 placed at 14:30 UTC against the current Pinnacle close" | clv_grade |
| "Price a same-game parlay: Yankees ML + Over 8.5 + Aaron Judge HR" | sgp_price |
| "Source-quality check: which books are showing degraded SLA right now?" | source_quality |
| "For each MLB game tonight, find the +EV play and write me a Python script that places it across two books" | find_ev + odds (multi-tool chain) |
If your stack is GPT-4, Gemini, Llama, or an OpenAI-compatible local model, you can still get the same agent shape. Two paths:
ParlayAPI ships a full OpenAPI 3.1 spec at /openapi.json. Most function-calling frameworks (OpenAI tools, Anthropic tools, Gemini function-calling, LangChain, LlamaIndex) can ingest OpenAPI specs directly and expose each endpoint as a callable tool to your model.
# OpenAI Python SDK example
from openai import OpenAI
import requests, json
tools = [] # auto-generate from /openapi.json or define manually
# ... your standard function-calling loop calls our endpoints ...For more control, give your model the curl cookbook as context and let it construct calls directly. Works for any model that can generate JSON or curl commands.
Cron job that pulls top +EV plays across MLB / NBA / NFL each morning and emails a digest. ~40 LOC.
Polls the full arb endpoint, posts to Slack when a new ML arb opens with edge ≥ 1%. ~25 LOC.
Overnight batch-grade yesterday's bets against today's morning closing lines. Idempotent. ~30 LOC.
Combine prop legs into 2-leg SGPs, price them, flag combinations that price softer than the independent product. ~50 LOC.
Detect when Pinnacle moneylines move more than 5 cents from the prior snapshot. Useful for "steam move" alerts. ~25 LOC.
Page on-call when any of your critical books goes "breach" or "stale". Cron every 5 minutes. No auth required. ~15 LOC.
code, message, and request_id the agent can surface back to the user. Every list response has consistent pagination. LLMs do dramatically better on APIs that are predictable than on APIs that are clever.
/v1/try/{sport}/* family lets your agent experiment without ever asking the user for a key. Easy onboarding inside an MCP-less LLM chat too.Free tier: 1,000 requests/month, no credit card. Paid starts at $5/mo (Starter) for 20k requests, $20/mo (Pro) for 100k. Full breakdown at /pricing. The MCP server itself is free and MIT-licensed, published at pypi.org/project/parlayapi-mcp.