Oracle3 is a free, open-source trading engine and MCP server for prediction markets. It is built for developers, quantitative researchers and anyone who wants AI agents to study contracts on Kalshi and Polymarket, and it is available for Windows as a ready-made build or installer. The software connects agents to those venues through the Model Context Protocol, so they can inspect rules, order books and related outcomes in a structured way.
Oracle3 is released under the Apache 2.0 license, and the current release is v1.2.2. Its source code is public on GitHub, and the latest code update is dated October 2026, which shows recent development activity. Because the program can place live orders as well as simulated ones, it deserves careful setup and a cautious approach, which this page covers in detail below.
What Is Oracle3 and Who Is It For?
Oracle3 sits between an AI assistant and a prediction market. Prediction markets let people trade contracts that pay out depending on whether an event happens, and platforms such as Kalshi and Polymarket list many such contracts. Oracle3 is designed for people who want to build research and trading logic on top of that data rather than trade by hand.
Typical users include Python developers who want to test ideas, researchers who study how related contracts are priced against each other, and technically inclined traders who prefer code over clicking. The project is a weaker fit for someone looking for a plug-and-play app with a dashboard, because the documented way in is through MCP clients and Python code.
If your goal is to learn how these markets behave before risking money, paper mode is the natural starting point. Treat it as a laboratory: run the same strategy against simulated fills, compare the results with what you expected, and only then consider live execution.
Real-World Scenarios for Agent-Driven Research
Consider a researcher who follows elections or economic data releases. They can ask an agent to list contracts on one topic, read the rules of each, and pull the current order book. The agent summarizes which outcomes are related and where prices look inconsistent, and the researcher decides which ones deserve a closer look.
A developer might take a different route. They write a strategy against the Python interfaces, run it in paper mode over several weeks of market snapshots, and review the simulated positions. Because the fee-aware view reports per-leg costs and net edge, the developer can see whether a strategy's gains survive realistic trading costs.
A small team may use Skills to keep the process consistent. A shared research routine, such as checking rules, comparing order books and logging each candidate basket, makes it easier to review decisions later. The value lies in repeatability and transparency, not in any promise about outcomes.
Key features
Research Workflow Built Around Contracts
Fee-Aware Relationship Checks
MCP and Skills for Agents
Extensible Python Engine
Reproducible Tests and Validation
Supported coins
Oracle3 is built for two prediction market venues, Kalshi and Polymarket, and its summary also covers Solana, with both paper and live execution available. The project documentation mentions Solana (SOL) among its supported coins and networks. Ledger is the hardware wallet named in the project materials, so it is worth reviewing the documentation for the exact integration you plan to use before connecting any account.
How to install Oracle3 on Windows
- 1
Get the program
Download the Windows build or installer from the official website or the GitHub Releases page. Choose the current release, which is v1.2.2 at the time of writing. Avoid copies from third-party download sites, because a modified file can contain code you did not expect.
- 2
Verify the file
Compare the SHA-256 checksum of the download with the value published alongside the release, or check the signature if one is provided. If the values do not match, do not run the file. Download it again from the official source.
- 3
Install into a dedicated folder
Run the installer or extract the build to a folder your Windows account can write to. Keep Oracle3 separate from other trading tools so that configuration files and logs do not get mixed up.
- 4
Create the virtual environment
The MCP server runs from a virtual environment. In the project folder, create a .venv and install the package into it, following the steps in the project documentation.
- 5
Note the server paths
On Windows the Python interpreter is .venv\Scripts\python.exe and the server executable is .venv\Scripts\oracle3-mcp.exe. Record the absolute paths, because your MCP client needs them.
- 6
Configure your MCP client
In your MCP client, set the command to the absolute path of oracle3-mcp.exe and leave the argument list empty. Restart the client so it loads the new server, then confirm that it lists the available tools.
-
Start in paper mode
Test the setup with simulated trading before adding any live credentials. Confirm that positions are tracked as expected and that fee and edge figures look reasonable.
System requirements
- Windows, listed as a supported platform through the Windows build or installer
- A Python installation that can create a virtual environment, because the MCP server is a command-line tool installed with pip into a .venv folder
- An MCP-compatible client if you want AI agents to call Oracle3 tools
- An internet connection for market data and for downloading the release
How to use Oracle3
Once the server is running, your first session should be a small, deliberate research task rather than a trading run. Pick one event you understand well and keep the scope narrow. Clear goals make it easier to judge whether the output is useful.
- Ask the agent to find contracts related to that event.
- Have it read the rules and current order book for each contract.
- Ask it to group related outcomes and compare their prices.
- Run one of the relationship checks and review the candidate basket.
- Read the per-leg fees and the calculated net edge before anything else.
- Track the idea as a simulated position in paper mode.
- Review the results after several sessions and change one variable at a time.
Keep a written log of each run so you can see why a decision was made. Once a routine works, save it as a Skill so the process stays consistent from session to session.
Pros & cons
- Free and open source under the Apache 2.0 license, so the code can be read and reviewed
- Fee-aware checks report per-leg fees and net edge instead of raw price gaps
- MCP and Skills give AI agents structured, repeatable access to research tasks
- Paper and live execution share one Python engine
- Automated tests and installed-package validation support reproducible work
- Available as a Windows build or installer
- Setup involves a virtual environment and command-line configuration, which is a hurdle for non-developers
- Focused on prediction markets, so it is not a general-purpose crypto trading bot
- Live execution involves real financial risk, and the engine does not remove model risk from statistical strategies
- The documentation is the primary guide, so expect to read it carefully before your first run
Is Oracle3 safe?
Oracle3 can place live orders and works with exchange credentials, so treat it with the same care as any trading software. Download it only from the official website or the GitHub Releases page, then verify the SHA-256 checksum or signature before running the installer. Because the code is public under Apache 2.0, you can read the source or build it yourself if you prefer, though building brings its own dependencies. Use API keys with the smallest permissions you need and start in paper mode. Never enter a seed phrase on third-party websites, and keep any Ledger hardware wallet under your own control. The project has recent code activity, but check the latest version on the official website before installing.
- Official source only. Download from github.com/YichengYang-Ethan/oracle3-prediction-market-agent. Ignore ads and lookalike domains.
- Check the SHA-256. Compare the installer hash with the value the developer publishes before running it.
- Never share your seed phrase. No developer, support agent or website — including CryptApps — will ever ask for it.
Alternatives
Oracle3 focuses on prediction markets on Kalshi and Polymarket, and its main interfaces are MCP and Python rather than a graphical dashboard. OctoBot and Gunbot cover more conventional trading bot workflows, with OctoBot offering a web interface. Jesse and Elinesoft XTester emphasize strategy frameworks and backtesting on historical data. TestMyLogic is closest to Oracle3 in spirit, since it checks strategies for overfitting. If you want an AI agent in the research loop and fee-aware checks across related contracts, Oracle3 is the more specialized choice.
Specifications
- Price
- Free
- Type
- Open source
- License
- Apache 2.0
- Latest version
- v1.2.2
- Last update
- 11 Oct 2026
- Windows support
- Yes (Windows build or installer)
- Source code
- github.com/YichengYang-Ethan/oracle3-prediction-market-agent
Frequently asked questions
Yes, Oracle3 is free and released under the Apache 2.0 license. The source code is public on GitHub, so you can read it and see how it works. Keep in mind that live trading on Kalshi or Polymarket involves real market risk and any venue fees, which the software cannot remove.
Yes, Oracle3 supports Windows through a Windows build or installer. After installation, the MCP server runs from a .venv folder, and the relevant files on Windows are .venv\Scripts\python.exe and .venv\Scripts\oracle3-mcp.exe. Use absolute paths when you configure an MCP client so it can find the server reliably.
Oracle3 can be used safely if you take basic precautions. Download it only from the official website or GitHub Releases, verify the SHA-256 checksum or signature, and begin with paper trading. Never share a seed phrase, and limit API key permissions to what you actually need.
Paper execution simulates orders and tracks positions without sending real trades to a venue. Live execution submits orders to supported venues using real funds. Use paper mode to test whether a strategy's logic behaves as you expect before any live capital is involved.
MCP, or Model Context Protocol, is a standard way for AI agents to call tools and read structured data. Oracle3 uses it so an agent can discover contracts, read order books and run research tools without custom integration code for each task. Any MCP-compatible client can connect once it knows the server command.
It means checking whether the prices of related contracts remain consistent after trading costs are included. If outcomes that must relate to each other are priced inconsistently, a possible gap exists. Oracle3 subtracts per-leg fees and reports the net edge, so a gap that disappears after costs becomes visible.
Yes, the project covers Solana alongside Kalshi and Polymarket for paper and live execution, and its documentation mentions Solana (SOL). Confirm the exact integration details in the documentation before you connect an account, since supported features can change between releases.
No. Oracle3 is a research and trading engine, and it does not promise returns. Net edge estimates depend on prices, fees and market conditions that change quickly. Treat paper results as learning material rather than evidence of how a strategy will perform in the future.
Verdict
Oracle3 suits developers and quantitative researchers who want AI agents involved in prediction market research, with fee-aware checks and paper trading built in. It is free, open source and available for Windows, but it asks for comfort with Python environments and MCP configuration. If you need a graphical bot for everyday exchange trading, OctoBot or Gunbot may be a better fit. Start in paper mode, read the documentation closely and verify each release before trusting it with real funds.