Polymarket for AI Agents — Live Prediction-Market Odds via MCP

When an AI agent gets asked “what are the odds of X?”, the right answer almost never lives in its training data. Prediction-market odds move in real time as new information arrives — by the time a model’s weights are baked, the odds it has memorized are months stale and probably wrong.

Polymarket is the largest crypto-collateralized prediction market in the world. Around 10,000 markets are open at any given time, spanning politics, crypto, sports, macroeconomic events, weather, scientific outcomes, and pop culture. Every market resolves to a Yes/No price in [0, 1] that you can read as an implied probability — and unlike polls or commentary, it’s backed by actual money so it incorporates information faster than almost any other public signal.

Pipeworx exposes Polymarket through six MCP tools, all live, all keyless. They’re the single most-called pack on the gateway by usage volume — agents reach for them for everything from “is Bitcoin going to hit $150k by June?” to “who’s leading in the Democratic primary?” to “what does the market think about tomorrow’s Fed meeting?”.

The tools

polymarket_search

Free-text search across active events. The first stop when the agent has a question but doesn’t know which Polymarket market addresses it.

polymarket_search(query: "rate cut", limit: 5)
→ [
    { event_title: "Will the Fed cut rates in June 2026?",
      markets: [
        { question: "...", yes_price: 0.62, volume_24h: 18230, ... },
        ...
      ]
    },
    ...
  ]

The yes_price is the implied probability. 0.62 = the market thinks there’s a 62% chance.

polymarket_market

Full detail for one specific market by slug or ID — including the resolution criteria text. Read this before quoting odds. A market called “Will Bitcoin hit $150k by June 30, 2026?” might resolve based on a specific exchange’s close price at a specific hour; the resolution text spells out exactly what “Yes” means.

polymarket_event

Polymarket groups mutually-exclusive markets into events. “2028 Democratic nominee” is one event with one Yes/No market per candidate. Use polymarket_event when you want every option at once — the full slate of candidates / outcomes / contenders with prices side-by-side.

polymarket_price_history

Historical probability time-series for one market. Returns [{timestamp, price}, ...]. Use to chart odds over time, detect probability jumps around news events (“the price moved from 0.30 to 0.65 in 90 minutes when X was announced”), or build backtests of market-implied vs realized probabilities.

polymarket_orderbook

Live CLOB orderbook — bid/ask ladders on both Yes and No sides with size at each price level. The liquidity field on polymarket_market is a rolled-up summary; this returns the actual ladder you’d hit if you wanted to size a position. Necessary before quoting numbers like “you could buy $50k of Yes at 0.62” — without checking depth, that’s just a guess.

polymarket_top_markets

Sorted by trading volume in a chosen window (24h / 1wk / 1mo / 1yr / all). The “what’s the market focused on right now” lens.

Patterns that work well

Implied probability vs your model. Ask the agent to compare Polymarket’s price against an internal estimate. If polymarket_market returns yes_price 0.18 on “Candidate X wins” and your model puts it at 0.05 based on polling, either the market knows something your model doesn’t or vice versa — and both directions are interesting.

Event-time probability shifts. Call polymarket_price_history around a known event timestamp (debate, earnings call, regulatory decision) to quantify how the market re-priced. Useful for “did this news actually move the market?” analyses.

Cross-venue spreads. Polymarket lists USD/crypto-denominated markets. Kalshi (also wired into Pipeworx as the kalshi pack) is the only CFTC-regulated US prediction-market exchange. For overlapping markets, the spread between their prices is informative — small spread means the market is liquid and efficient, big spread can signal mispricing or one venue lagging news.

Compound research. Polymarket prices are one input among many. Pipeworx’s compound bet_research meta-tool combines a Polymarket market lookup with relevant news (gdelt), historical analogues (wikipedia), and source-of-truth data (e.g. fred for macro markets) into one structured packet — useful when the agent needs to defend a probability estimate rather than just quote one.

A note on regulation

Polymarket is unavailable to US persons under current CFTC rules. Reading the odds is fine — anyone with an internet connection can do that — but trading requires being outside the US. If your agent’s task is “tell me the current odds,” Polymarket is the deepest source. If the task involves placing actual bets for a US-based user, route to kalshi instead.

Pipeworx returns the data; what the agent does with it is up to the application layer.