Skip to main content
Guide

Use PatternSight in your AI assistant

The MCP server lets Claude, ChatGPT, and other AI assistants use PatternSight directly — live results, statistics, calculators, and the prediction swarm — just by asking in plain English. You don't need to be a developer; connect it once and chat.

Live now. The MCP endpoint is connectable at https://api.patternsight.app/mcp/. Prefer code? The REST API exposes the same tools.

Connect in 3 steps

  1. 1

    Get your API key

    Open Dashboard → API Keys and click Create key. Copy it right away — it's shown only once. Keep it private, like a password.

  2. 2

    Connect it to Claude Desktop

    In Claude Desktop: Settings → Developer → Edit Config. Paste the block below (or merge it into your existing mcpServers), replacing YOUR_API_KEY with the key from step 1. Save and restart Claude.

    {
  "mcpServers": {
    "patternsight": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://api.patternsight.app/mcp/",
        "--header", "Authorization: Bearer YOUR_API_KEY"
      ]
    }
  }
}

    Uses mcp-remote (npx fetches it automatically; it needs Node.js installed). Other MCP clients: point them at https://api.patternsight.app/mcp/ with the header Authorization: Bearer YOUR_API_KEY. ChatGPT connects over OAuth — a separate, later release.

  3. 3

    Ask in plain English

    Once connected, just ask. For example:

    • “What were the most recent Powerball numbers?”
    • “Show the hottest and coldest Mega Millions numbers over the last 5 years.”
    • “Which Powerball numbers have gone the longest without being drawn?”
    • “What are the odds of winning EuroMillions, and the after-tax value of a $20M jackpot?”
    • “Generate 5 balanced Powerball picks, then backtest a balanced strategy. (paid tier)”

    Your assistant decides which tools to call and explains the results. Everything is descriptive analysis of past draws.

Prefer to call it from code?

Every tool is also a plain REST endpoint — no MCP client needed. Send your key as a Bearer token:

curl https://api.patternsight.app/v1/games/powerball/frequency \
  -H "Authorization: Bearer $PATTERNSIGHT_API_KEY"

See the full REST API reference and authentication guide.

Endpoint

Streamable HTTP at https://api.patternsight.app/mcp/. Works with Claude, direct MCP clients, and the OpenAI Responses API.

Auth

Authorization: Bearer with your PatternSight API key. Get one in Dashboard → API Keys.

Entitlement

Data & calculator tools work with any key. generate_picks and backtest need a paid tier — Pro or Ultra.

All tools

list_gamesmeter: api_requests

List available lottery games and whether picks/backtest are supported for each.

get_gamemeter: api_requests

Get one game's number configuration (pool, range, special ball).

latest_resultsmeter: api_requests

Most recent draw result(s) for a game. Past results only — descriptive.

timeseriesmeter: api_requests

Historical draw series for a game (dates, numbers, sum/odd composition).

frequencymeter: api_requests

Number-frequency analysis with a chi-squared residual.

Describes past draws.

gapsmeter: api_requests

Gap analysis: draws since each number last appeared, with mean/median/max gap.

Draws since each number last appeared.

composition_analysismeter: api_requests

Odd/even, high/low, sum, and band composition of past draws vs a fair draw.

analyze_gamemeter: api_requests

Composed descriptive analysis — frequency, gaps, and draw uniformity from precomputed digests.

odds · expected_value · estimate_tax · lump_sum_vs_annuity · wheeling · syndicate_splitmeter: api_requests

The full calculator suite — jackpot odds, expected value, after-tax winnings, lump sum vs annuity, wheeling coverage, and syndicate splits.

generate_picksmeter: agentic_analysesPaid tier — Pro or Ultra

Generate number picks using the swarm's mathematical strategies. Each result carries a provenance pack_hash and the disclaimer.

Mathematical number selections.

backtestmeter: agentic_analysesPaid tier — Pro or Ultra

Walk-forward backtest of a strategy against historical draws, with no lookahead.

edge_vs_random clusters near 0 — that means no real edge. A small non-zero value is sample noise, not a winning signal: no strategy beats random on fair lottery draws.