Finance MCP servers with inspectable evidence

Choose the task you need to answer

Combine the servers in a research workflow

Start with the research server to inspect a mechanism, its failed candidates and the declared search history. Use fundamentals to reconstruct the accounting information available by a decision date. After you construct strategy returns with your own causality and cost controls, use validation to assess selection bias, overfitting and record maturity.

These stages answer different questions. Accounting facts do not become returns automatically, published papers can contain failed experiments, and statistical tests do not certify a strategy's future profitability.

Source data, datasets and reproduction

• SEC company reference and source filings
• FilingFacts dataset: financial questions with source evidence
• FilingFacts review and annotation workspace
• Null Zoo research benchmark
• Open paper evidence reporting standard
• ALPHAC, point-in-time lake and backtester repositories

FilingFacts v0 is machine-checked; its review packet does not establish independently adjudicated human gold. The paper evidence standard is a proposal with disclosed implementation and review limits.

Release, privacy and execution scope

Hosted endpoints import pinned released source. npm distribution can lag those pins, so each server page explains which hosted contract its tool list describes. Read the setup guide before choosing remote or local validation and before sending private strategy inputs.

The research and fundamentals servers read public records. Remote validation stores reproducible receipts under its free-key quotas. The execution package under development has local planning and journal tools; it is private and unreleased, and this directory does not advertise a hosted trading endpoint or live broker service. The execution-cost calculator is available in the browser. Download the discovery record with the release pins and examples.