Back

Prediction Pools MCP

Tool specifications — The pool tools below are exposed through the main EventTrader MCP server. All are read tools and work unauthenticated.

The Model Context Protocol (MCP) server exposes the prediction pools to LLM clients (Claude Desktop, Claude Code, MCP-compatible agents) — the BTC pools (btc-4h, btc-5m) and the tradable-event pool on the landing page (event-5m, whose index is the price-weighted basket of the assets moved by the featured news headline; its state carries the headline card as subject). Ask your AI assistant where the live prediction tape is clustered, what the last settlement price was, or how big the current pool is — all through natural language.

REST & WebSocket Python SDK MCP Tools 5-Minute Pool 4-Hour Pool

Server Endpoint

https://cymetica.com/mcp/v1

EventTrader exposes a single MCP surface — the same endpoint serves arena, exchange, OTC stocks, TGE-launch, Horse Race, and the BTC prediction pools. The tools below filter to the BTC pools when called.

Discovery: GET /.well-known/mcp.json · Tools: GET /mcp/v1/tools · Register: POST /mcp/v1/register

For Claude Desktop, add to claude_desktop_config.json:

{
  "mcpServers": {
    "eventtrader": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://cymetica.com/mcp/v1"]
    }
  }
}

Tools

Tool	Args	Description
`get_btc_pool_state`	`pool?`	Full public state of one pool (`btc-4h`, `btc-5m`, or `event-5m`): open entry round with the anonymized prediction tape, locked measuring round, last settled/voided round, live index (BTC/USD, or the event index + headline `subject` for `event-5m`).
`get_btc_pool_history`	`pool?, limit?`	Recent terminal rounds newest-first: P0/P1 snapshots, winning price and distance, winners count, pool size, void reasons.
`get_prediction_statistics`	`pool?, round?, history_window?`	Aggregated crowd analytics for one round: consensus price (stake-weighted median) and its distance from the live index, mean/median, standard deviation, dispersion percent, 0–1 consensus score, bullish/bearish split by entries and stake, quartiles, largest prediction cluster, and the historical percentile of the current spread.

Authentication

All three tools are read-only and work unauthenticated. Entering and cancelling move money, so they are not exposed as MCP tools — use the authenticated REST endpoints POST /api/v1/btc-pool/enter and POST /api/v1/btc-pool/cancel (JWT) or the Python SDK.

Example LLM Prompts

"What's the state of the 5-minute BTC pool right now?"
"Where are predictions clustered in the current round — above or below the index?"
"What did the last 10 rounds of the btc-5m pool settle at?"
"How often does the winning distance come in under $5?"
"How big is the current 4-hour pool and when does it lock?"
"Is the crowd unusually bullish this round?"
"How concentrated are predictions right now?"
"Is today's prediction distribution wider than average?"

Tool Reference

get_btc_pool_state

{
  "name": "get_btc_pool_state",
  "description": "Public state of a BTC prediction pool. Returns the open entry round (with the anonymized prediction tape), the locked measuring round, the last settled/voided round, and the live BTC/USD index.",
  "inputSchema": {"type":"object","properties":{
    "pool":{"type":"string","enum":["btc-4h","btc-5m"],"default":"btc-4h"}
  }}
}

get_btc_pool_history

{
  "name": "get_btc_pool_history",
  "description": "Recent terminal rounds (settled or voided) of a BTC prediction pool, newest first — P0/P1 snapshots, winning price/distance, winners count, pool size.",
  "inputSchema": {"type":"object","properties":{
    "pool":{"type":"string","enum":["btc-4h","btc-5m"],"default":"btc-4h"},
    "limit":{"type":"integer","minimum":1,"maximum":200,"default":24}
  }}
}

get_prediction_statistics

{
  "name": "get_prediction_statistics",
  "description": "Aggregated crowd analytics for ONE round — consensus price, dispersion, sentiment split, largest cluster, historical percentile. Read-only.",
  "inputSchema": {"type":"object","properties":{
    "pool":{"type":"string","enum":["btc-4h","btc-5m","event-5m"],"default":"btc-4h"},
    "round":{"type":"string","enum":["entry","measuring","last"],"default":"entry"},
    "history_window":{"type":"integer","minimum":1,"maximum":500,"default":100}
  }}
}

Saves every agent from reconstructing the same statistics out of the raw prediction tape. consensus_price is the stake-weighted median; consensus_score is 0–1 (1 = unanimous). dispersion_vs_typical ranks the current spread against recent completed rounds, answering “is today’s distribution wider than average?”. Bullish/bearish is derived from predicted price vs the round’s reference price (lock price once locked, otherwise the live index), never from a stored side column. Rounds with fewer than 10 predictions set low_sample: true and return null for distribution-shaped fields rather than emitting precise-looking numbers from a handful of entries.

Limits

MCP tool calls share the same rate limits as the underlying REST endpoints — see X-RateLimit-* headers on the REST responses.

EventTrader