MCP Hub
Back to servers

mcp-server

Prediction market probability oracle. 500+ markets from Kalshi, Polymarket & Metaculus.

Registryglama
Forks
1
Updated
Mar 19, 2026

Quick Install

npx -y telekash-mcp-server

TeleKash Prediction Oracle — MCP Server

The probability oracle for the agent economy. 13 tools across 500+ live markets from Kalshi, Polymarket, and Metaculus.

"Chainlink is the price oracle. TeleKash is the probability oracle."

Installation

npx telekash-mcp-server

Or add to Claude Code:

claude mcp add telekash-oracle npx telekash-mcp-server

Or add to your MCP config:

{
  "mcpServers": {
    "telekash-oracle": {
      "command": "npx",
      "args": ["telekash-mcp-server"],
      "env": {
        "SUPABASE_URL": "https://rrkjtdnxkscukexbsrue.supabase.co",
        "SUPABASE_ANON_KEY": "your-key"
      }
    }
  }
}

Tools

13 oracle tools querying 500+ live markets from 3 sources, synced every 15 minutes.

Intelligence

ToolWhat It Does
get_probabilityReal-time YES/NO probability with volume-weighted confidence score
list_marketsBrowse markets by category (crypto, politics, sports, economics, weather)
search_marketsFull-text search across all markets
get_historyHistorical probability changes with trend detection
get_sentimentAI sentiment analysis — conviction, momentum, volume, noise filter
get_market_statsAggregate statistics across all markets
get_trendingMarkets with biggest probability swings — signal vs noise detection
compare_sourcesKalshi vs Polymarket vs Metaculus odds comparison

Signals

ToolWhat It Does
get_signalStructured TPF signal — one call for complete trade decision (STRONG_BUY → NO_SIGNAL)
detect_arbitrageCross-source arbitrage detection with buy/sell signals
get_divergencesConsensus divergence detection — where sources disagree most

Agent Performance

ToolWhat It Does
track_predictionRecord a prediction for accuracy tracking
get_performanceBrier score, calibration curve, edge-vs-market analysis

TeleKash Probability Format (TPF)

The get_signal tool returns a structured signal combining all intelligence into one actionable verdict:

get_signal(query: "Will Bitcoin hit 100k?")
→ {
    probability: { yes: 72, confidence: { grade: "HIGH" } },
    sentiment: { recommendation: "bullish" },
    noise_filter: { signal_quality: "signal" },
    cross_source: { spread_pct: 4.2 },
    verdict: { action: "BUY", score: 28.5 }
  }

One call replaces get_probability + get_sentiment + get_history + compare_sources.

Examples

See the examples/ directory for complete agent scripts:

  • portfolio-scanner.ts — Scan markets, find high-confidence signals, build ranked portfolio
  • arbitrage-hunter.ts — Monitor cross-source spreads and alert on opportunities
  • market-monitor.ts — Watch a specific market and log probability changes

Quick prompts for your AI agent:

  • "What are the odds Trump wins 2028?"get_probability
  • "Find arbitrage opportunities"detect_arbitrage
  • "Where do prediction sources disagree most?"get_divergences
  • "Give me a trading signal for Bitcoin markets"get_signal
  • "Track my prediction: YES on Bitcoin >100k at 75%"track_prediction
  • "How accurate are my predictions?"get_performance

Data Sources

SourceTypeSync Frequency
KalshiCFTC-regulated US exchangeEvery 15 min
PolymarketCrypto-native exchangeEvery hour
MetaculusForecaster consensusEvery 2 hours

All sources aggregated, deduplicated, and categorized automatically.

Unique Capabilities

Noise detection — Serial correlation reversal analysis on 15-min probability snapshots. Classifies momentum as signal/weak/noise. "58% of price moves are noise."

Consensus divergence — Finds where Kalshi, Polymarket, and Metaculus disagree. STRONG divergences (>15%) mean at least one source is significantly wrong — that's where alpha lives.

Agent accuracy tracking — Brier scores, calibration curves, and edge-vs-market analysis. Build a verifiable prediction track record.

Volume-weighted confidence — Every probability includes a confidence grade (HIGH/MEDIUM/LOW/VERY_LOW) based on volume, liquidity, conviction, and time decay.

Cross-source arbitrage — $40M+ extracted from prediction market mispricings annually. This tool finds them automatically.

Environment Variables

Works without credentials (returns demo data). For live markets:

SUPABASE_URL=https://rrkjtdnxkscukexbsrue.supabase.co
SUPABASE_ANON_KEY=your-anon-key

Links

License

MIT

Reviews

No reviews yet

Sign in to write a review