Bring recorded market evidence into your workflow.

Ask a question. Review the study. Read the evidence.

Describe the setup you want to investigate. Your assistant can discover available readings, propose an explicit definition and ask you to approve the study before it runs. The result shows how much evidence exists, what followed and the limitations that matter.

For example: “When open interest rose while liquidations were elevated, what followed over the next hour?” Review the actual thresholds, markets and dates before accepting the proposed interpretation. Available fields and coverage determine what can be tested.

MCP: research inside your assistant

Connect a compatible MCP client to https://mcp.edgedepth.com/mcp and authorize through the hosted setup. The MCP guide covers supported authentication and local stdio configuration.

MCP supports study preparation, approved scans, historical inspection and report retrieval. Compatible hosts can display an inline result comparison and distribution. The underlying result remains available for inspection even when the host does not support the visual interface.

Saving and alert management happen in the web app. The MCP research tools do not place trades or change alerts. Follow the result back to the Workbench to save a study or confirm monitoring of a supported definition.

REST: exact documents and machine-readable results

Use the registry to discover supported fields and operators, then submit a versioned query document. The API returns canonical engine JSON with the query hash, dataset revision, counts and forward-outcome summaries.

Keep the result and its metadata with your analysis. A page of returned matches is a set of examples; whole-result summaries describe the matching population. Cursor pagination does not turn the current page into the denominator.

Create an API key, then follow the registry-to-result quickstart. Keep bearer keys on trusted systems, outside public browser code.

Evidence your workflow can check

• Explicit scope. Inspect the conditions, markets, dates and measurement definitions before computation.
• Counts and references. Read the outcome denominator and missing-window count. The unconditional same-scope reference includes matches; it is not a matched control group.
• Separate selection from outcomes. Setup studies measure what followed their anchors. Outcome-first discovery is available as a separate retrospective workflow and cannot establish predictive performance.
• Revision-aware identity. The exact query and data revision identify a result. New data or revised definitions can produce a new identity and different results.
• Evidence handoffs. Open supporting and contradictory examples in the browser, with tick replay where recorded coverage and access allow.

Coverage and access

Searchable research currently covers Binance USDT-M. Live Hyperliquid coverage is separate from searchable history. Check market and feature coverage before choosing a window.

Browser, API and MCP use account research allowances. New computation can consume credits; cache hits and unchanged ETag revalidations do not. Broad or longer studies can cost more than one unit. Reading an existing public report is unlimited. See credits and limits and plan access.

Build further, or start in the browser

The open-source terminal, gateway and MCP server expose different parts of the stack. Self-hosting the terminal does not include the hosted research corpus. For bulk data or licensing, see data for teams.

Prefer to explore a question first? Open the Workbench or read a published study. Historical evidence helps you choose a next check; it does not establish profitability or causality.