An arbitrage engine that remembers the fees
Oracle3 wires LLM agents into Kalshi and Polymarket with MCP tools that check no-arbitrage relations between contracts — net of fees, not just on paper napkins.
What it does
Oracle3 is a Python trading engine plus MCP server for prediction markets. It discovers contracts on Kalshi and Polymarket, inspects their rules and order books, and checks five logical relationships between related contracts — implication, exclusivity, complement, same event, and event sum. When a relationship is violated, it builds a candidate arbitrage basket and reports gross edge, per-leg fees, and net edge. Execution can be simulated in a local paper ledger or run live via a separate CLI/Python engine with pre-trade risk limits.
The interesting bit
Most arbitrage talk ignores fees until the trade evaporates. Oracle3 makes fee-adjusted net edge the default output of every calculation, and it exposes the assumptions behind each relationship check — so an agent (or a human) can see exactly what a basket is conditional on. The synthetic example in the docs is refreshingly honest: $5.00 gross edge shrinks to $1.59 after fees.
Key highlights
- 13 MCP tools covering market discovery, quotes, order books, fee math, fair-value models, and paper trading
- Six reusable agent Skills spanning research, strategy authoring, and paper/live operations
- Public market reads need no exchange credentials; live execution is deliberately split off from the MCP layer
- Trading engine with paper/live adapters for Kalshi, Polymarket, and Solana via DFlow, plus Wang-transform pricing models
- CI validates the built package through real MCP sessions on SDK 1.x and 2.x; 600+ tests per the repo description
Caveats
- The README is explicit that this is research and education software, with the usual financial-risk disclaimer
- Live trading requires the separate engine and credentials — the agent-facing MCP layer is read/paper only by design
- Model-level agent evaluation is listed as a “next research direction,” so the agent side is still maturing
Verdict
Worth a look if you build trading agents or want a fee-honest calculation layer for event contracts. If you wanted a turnkey money-printing bot, this is a research harness, not a promise.
Frequently asked
- What is YichengYang-Ethan/oracle3-prediction-market-agent?
- Oracle3 wires LLM agents into Kalshi and Polymarket with MCP tools that check no-arbitrage relations between contracts — net of fees, not just on paper napkins.
- Is oracle3-prediction-market-agent open source?
- Yes — YichengYang-Ethan/oracle3-prediction-market-agent is open source, released under the Apache-2.0 license.
- What language is oracle3-prediction-market-agent written in?
- YichengYang-Ethan/oracle3-prediction-market-agent is primarily written in Python.
- How popular is oracle3-prediction-market-agent?
- YichengYang-Ethan/oracle3-prediction-market-agent has 507 stars on GitHub.
- Where can I find oracle3-prediction-market-agent?
- YichengYang-Ethan/oracle3-prediction-market-agent is on GitHub at https://github.com/YichengYang-Ethan/oracle3-prediction-market-agent.