parlayapi-mcp: ParlayAPI as native MCP tools for Claude / Cursor / any AI agent

Install in 30 seconds

Quickest path - no virtualenv, no clone:

uvx parlayapi-mcp --help

Or with pip:

pip install parlayapi-mcp

Both fetch the package from PyPI on first run. Current version: 0.3.5. The wheel is small and pulls in the official MCP SDK and httpx as dependencies. It is also listed in the official MCP registry as io.github.JacobiusMakes/parlayapi.

No install at all: remote endpoint

The same 22 tools are hosted at https://parlay-api.com/mcp/http over MCP streamable HTTP. Point any remote-capable client (claude.ai connectors, hosted agents) at that URL. Keyless requests get the discovery and signup tools; add Authorization: Bearer <api_key> for the full set.

Configure your MCP client

Claude Desktop (macOS)

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "parlayapi": {
      "command": "uvx",
      "args": ["parlayapi-mcp"],
      "env": {
        "PARLAYAPI_KEY": "your_key_here"
      }
    }
  }
}

Don't have a key yet? Omit env entirely. The parlayapi_signup tool works without a key; so do the public live preview, pricing, source-quality, and coverage tools. Add the returned key to env and restart Claude Desktop when you want paid data endpoints.

Cursor

Edit .cursor/mcp.json in your project (same shape).

OpenClaw

Save a keyless MCP definition first, then let the assistant prove live coverage before you add a key:

openclaw mcp set parlayapi '{"command":"uvx","args":["parlayapi-mcp"]}'

After signup, add the key:

openclaw mcp set parlayapi '{"command":"uvx","args":["parlayapi-mcp"],"env":{"PARLAYAPI_KEY":"your_key_here"}}'

Continue / Devin / custom MCP client

Any client that supports MCP stdio transport works. Run parlayapi-mcp as the command, set PARLAYAPI_KEY in env. The wire protocol is standard MCP and the tools auto-register on startup.

All 22 tools

Tool	Auth	Purpose
`parlayapi_signup`	none	Create a free-tier account, return API key + claim URL
`parlayapi_checkout_link`	none	Stripe upgrade URL for any tier
`parlayapi_magic_link`	none	Email passwordless login link
`parlayapi_get_pricing`	none	Current pricing and credit tiers
`parlayapi_live_sports`	none	Active live sports, event counts, and book counts
`parlayapi_live_search`	none	Search live teams, players, and sport keys
`parlayapi_live_command_center`	none	Public best-line preview from /live
`parlayapi_source_quality`	none	Source freshness and quality metadata
`parlayapi_book_coverage`	none	Per-book five-gate coverage proof
`parlayapi_list_sports`	key	90+ sport keys (MLB, NBA, NFL, NHL, EPL, UFC, esports, ...)
`parlayapi_get_odds`	key	h2h, spreads, totals across 30+ books for a sport
`parlayapi_get_props`	key	Player prop odds (DK, FanDuel, BetMGM, PrizePicks, Underdog, ...)
`parlayapi_best_line`	key	Best price per outcome across all bookmakers
`parlayapi_verdict`	key	"Should I bet this?" One-call verdict: no-vig fair price, best book, EV grade, BET / LEAN / FAIR / PASS
`parlayapi_parlay_verdict`	key	Grade a 2-12 leg parlay: combined fair price, best single book for the slip, weakest leg
`parlayapi_best_bets`	key	Ranked +EV plays for a sport, scoped to books the user can bet at
`parlayapi_set_bettable_books`	key	Remember the user's region or book list so verdicts shop only bettable books
`parlayapi_find_arbitrage`	key	Guaranteed-profit arbs across books, including 3-way soccer markets
`parlayapi_find_ev`	key	+EV bets vs a sharp book's no-vig fair line
`parlayapi_consensus`	key	Average, best, and worst price per market across all bookmakers
`parlayapi_find_middles`	key	Cross-book middles on totals, spreads, and player props, with per-$100 economics
`parlayapi_account_info`	key	Tier, credits remaining, billing period

Tools marked none do not require an API key (they create or manage one). Tools marked key need PARLAYAPI_KEY in the MCP client's env config for the parlayapi server.

What an agent flow looks like

User says to their assistant: "build me a discord bot that posts NBA +EV plays from sharp books in real time".

With parlayapi-mcp configured, the agent:

Calls parlayapi_signup(email). Gets back an API key and a magic-link claim URL. Pastes the claim URL to the user, drops the API key into the user's bot config.
Calls parlayapi_list_sports() to confirm basketball_nba is active.
Calls parlayapi_best_line(sport_key="basketball_nba", market="h2h") to see how the best-line endpoint shapes the data.
Writes the actual Discord bot code, references the API key, runs it locally for the user.

Total time from "I want a bot" to working bot: two minutes if the agent is fast. No browser tabs, no manual signup form, no copy-pasting an API key.

ParlayAPI as native MCP tools