EdgeDepth Research

EdgeDepth Research MCP Server

@edgedepth/research-mcp is the official, research-only Model Context Protocol server for EdgeDepth, a market microstructure search engine over recorded Binance USDT-M crypto perpetuals. Use it from ChatGPT, Claude, Cursor, Codex, or any MCP client to find every verified occurrence of a market condition, inspect forward outcomes across the complete matched set, read an unconditional same-scope reference, and open replay-linked evidence.

Every result includes counts with denominators and a reproducibility key. Same key, same bytes.

Website · Search the market · REST API documentation · MCP setup guide · Learning hub

Why use EdgeDepth Research?

• Search recorded market microstructure: query a closed, versioned feature registry covering order flow, order-book depth and cost to trade, capture provenance, price action, volatility, funding, open interest, positioning, candle formations, liquidations and liquidation shelves.
• Keep the denominator: every count reports the eligible population and exclusions behind it. Missing data is absent, never silently changed to zero.
• Measure outcomes without lookahead selection: forward returns, MFE, and MAE are computed over all occurrences. Outcome fields cannot be used as filters.
• Compare matched and baseline populations: deterministic cohort results put the matched distribution beside every other eligible predicate-false bucket.
• Audit and replay the evidence: results carry a reproducibility key, and representative occurrences include authenticated web handoffs to the exact recorded market moment.
• Stay research-only: no tool trades, modifies alerts or publishes reports. The private Radar admin pilot can save account-owned hypotheses with the separate optional research:hypotheses permission. A fresh scan, cohort, or stratified computation can consume research allowance units; the annotations state that side effect explicitly.

Choose a connection

The package exposes one tool core through two transports:

• Hosted MCP (recommended): connect to https://mcp.edgedepth.com/mcp over Streamable HTTP and authorize once in your browser. No API key to copy.
• Local stdio: run npx -y @edgedepth/research-mcp with an EdgeDepth API key.

Connect

Claude Desktop

In Settings > Connectors > Add custom connector, enter:

https://mcp.edgedepth.com/mcp

Complete the EdgeDepth browser authorization prompt.

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "edgedepth-research": {
      "url": "https://mcp.edgedepth.com/mcp"
    }
  }
}

Codex (~/.codex/config.toml)

[mcp_servers.edgedepth]
url = "https://mcp.edgedepth.com/mcp"

Then run:

codex mcp login edgedepth

Remove any old bearer_token_env_var line before using browser OAuth.

Local stdio with npx

Create a key on the EdgeDepth Developer page, then add:

{
  "mcpServers": {
    "edgedepth-research": {
      "command": "npx",
      "args": ["-y", "@edgedepth/research-mcp"],
      "env": {
        "EDGEDEPTH_API_KEY": "edk_live_YOUR_KEY"
      }
    }
  }
}

Local stdio requires Node.js 20 or newer. Use the research:read key scope for recorded-data tools and add research:interpret only when you need the free interpret_prose proposal step.

Result projection (agent context economy)

Scan-family results are large: a universe scan's canonical bytes run to hundreds of kilobytes, most of it page rows carrying every recorded feature, the zero and long-tail entries of counts_by_symbol, and empty threshold rungs. That overflows a client's tool-result budget before it answers anything.

run_scan, next_page and run_cohort therefore return a stated projection by default. Display derivations and removals are stated with the exact way to recover detail. Default run_scan adds the compact text view described below; full_outcomes: true, next_page and run_cohort keep this detailed projection:

• occurrence rows are trimmed to rows (default 3) and each kept row keeps the setup fields its own evidence block names - full_rows: true restores the whole vector;
• the per-occurrence outcomes map keeps the entries for the rows that remain;
• counts_by_symbol keeps the top entries by match count, and says how many instruments and matches were omitted;
• the outcome ladders are replaced by a paired answer block: for each metric, present, absent and the selected rungs' integer counts pass through verbatim, with rate, the unconditional baseline_rate over the same symbols and window, and their ratio as lift stated beside them. The selection is fixed in advance (gte 0.01, gte 0.02, lte -0.01, lte -0.02), drops rungs that separate nothing, and adds the single rung carrying the largest lift among those holding at least 30 occurrences, marked kept_for. full_outcomes: true returns every rung and the per-rung histogram, on the matched set and the reference separately.

Full-population counts, outcome denominators, absent tallies, predicate_coverage, representatives, the page cursor and reproducibility key stay intact. The request document is never rewritten, so the canonical query hash and the credit charged are exactly what you asked for. full_counts: true returns the engine's verbatim canonical bytes with no projection at all. ETags are projection-scoped: an ETag held for one projection can never revalidate as a different one.

list_features takes the same treatment on request: search, feature_ids and compact return one feature family instead of the whole grammar, with the closed parts (operators, windows, sequence rules, limits, error codes) intact.

Prompts and resources

The server publishes worked prompts, which compatible clients surface as pickable commands: test_a_claim, liquidation_cascade_bounce, investigate_symbol, what_preceded_moves_like_this, does_it_confirm and how_common_is_it (the free prevalence path). Each one encodes the same answer contract: ground the grammar, propose the exact definition, wait for confirmation, then report with denominators, the reference, the reproducibility key and a replay handoff.

The grammar registry is also served as a resource, edgedepth://research/grammar, so a client can attach it once instead of calling list_features every session.

Recommended agent workflow

1. For setup-first questions, interpret prose in the host and call prepare_study with structured scope, predicates and outcome. It reuses the web validator and measure contract, validates current instrument membership and returns an unrun canonical definition plus a fresh allowance estimate. No external LLM is used. Preserve user_stated, semantic_translation (top 10% means rank >= 0.9) and model_assumed separately; only the latter denotes an invented proposal. Use interpret_prose unchanged as the raw-prose fallback. Use compact list_features for uncommon fields or validation repair, not every question.
2. Call list_instruments only when you need to check the manifest-derived universe, coverage, and provenance. Its result carries the human market page in the same way, https://edgedepth.com/research/symbols/<symbol>, for a market still being recorded; a delisted market in the universe has no page, so offer that link rather than promising it.
3. Show one short proposal: condition, exact markets and dates/time zone, outcome definition and horizon, and metering. Interpretation is free; fresh computations can consume allowance. Label every unprovided value as a proposed assumption using chip provenance. Resolve unsupported fragments and ask only questions that materially change the study. Keep exact JSON and diagnostics inspectable in tool details, available on request.
4. Wait for explicit human approval, then pass the same document to run_scan. Changes require a new proposal and confirmation. The exact-document API does not store a proposal ID or a human approval receipt; client consent is required, and a model-supplied flag is not proof. On the supporting web release (b68c744 or later), returned rq workbench links load editable proposals and wait for Run; navigation never authorizes computation.
5. Answer the question first, preserving zero-match and inconclusive findings. Give matched/eligible counts, coverage exclusions, present/absent outcomes, both directions at the agreed horizon, and overlap/selection limitations. Read rates from outcomes_summary, which covers all occurrences. Page rows are examples, never the denominator. Each rung already carries its matched count and rate, the unconditional rate, and their ratio as lift: quote those, and quote the count beside the rate. No lift means no reference was available or the unconditional rate was zero; neither licenses estimating one.
6. Read the appended unconditional same-scope reference when available. It is not matched, comparable, or a causal control.
7. Return the full reproducibility key with the answer and one relevant next action: a returned replay, a changed assumption, or an existing report. General saved studies and alerts remain web actions; private Radar hypotheses use the optional workflow below. Each handoff states how far back it sits; replay reach is a per-account entitlement, so an old moment can be refused at the web surface even though the occurrence is real. Use next_page only with a cursor returned by the API.

Example instruction for an MCP client:

Did elevated VPIN and one-sided buying tend to precede a rise? Propose a precise
study before running anything. Label any suggested thresholds, markets, dates
and outcome definition so I can approve or change them.

The user does not need tool names, feature IDs or JSON. The client translates the confirmed proposal into the existing exact-document call.

Outcome-first and pointed-move workflow

For an outcome-first question, use outcome_first after agreeing the target and scope. Preserve touched-within (reached) versus close-at-end (finished), direction, size and horizon. Do not pass the outcome to the setup interpreter or substitute the worked example. The target grammar is available at edgedepth://research/outcome-first.

Report the population and both counted shares for each displayed reading. Help the person choose one reading, retrieve its setup_first_rerun with full_rows: true on the unchanged request, and confirm that exact setup before run_scan. Pass the original target as run_scan.measure outside the unchanged document: kind: "touch" for reached, "close" for finished, plus the agreed direction, fractional magnitude and horizon. The returned workbench link keeps that display choice and remains an unrun draft. This does not alter the scan/cache key. The local selected-outcome addition below preserves the exact reading separately from closing-return exploration. Read the original outcome target from the complete matched-set summary; request full_outcomes if the projection omitted its rung. An unavailable rung is stated, never replaced by the default horizon. The two reads have different denominators. A same-period rerun remains exploratory; freeze the condition and use a separate period before claiming validation.

A named moment can be inspected with snapshot_at; commonality compares multiple supplied moments. The screenshot path below adds bounded explicit close-range investigation and the existing detector geometry. Automatic move selection is not exposed through MCP. Historical marker browsing and general volume-tier resolution are not MCP capabilities yet. The local resolve_scope addition below supplies explicit sector resolution after its web release. list_instruments supplies coverage and instrument provenance, not sector membership. Use resolve_scope for recorded sector membership when available; otherwise use an exact supplied roster; never invent group members or a numeric price. Replay handoffs open the web surface and remain subject to the person's coverage and entitlement.

Tools

No tool can trade, change market state or publish. Only the private hypothesis tools modify account research history, with separate permission. run_scan, run_cohort, run_stratified and outcome_first are annotated as metered computations because a fresh call can irreversibly consume an allowance unit. The other recorded-data tools are closed-world reads. interpret_prose is a free read that uses the configured external language interpreter.

Private Radar hypothesis workflow (local addition)

Requires the supporting web release, its optional API-key scope migration and Radar admin access. Web precedes MCP. Existing grants and keys gain no permission; request research:hypotheses alongside research:read and authorize a new grant, or explicitly select the hypothesis option when creating a key. Ordinary accounts remain denied. This is not a public release or hosted acceptance claim.

• get_hypothesis: original observation (event), your saved record (id), or your list (neither). full: true restores all receipt bodies. Reads run nothing.
• prepare_hypothesis: exact editable draft, four requests, revision, plan hash and an allowance estimate. No save, computation or approval is implied.
• save_hypothesis: explicit private save at the known revision; existing source and attempts stay immutable. Save Test/Keep/Reject without dropping failures.
• run_hypothesis: after human approval of the exact proposal, freeze all four requests and run the first. Use the returned continuation serially for the remaining requests under that same approval. Stop when it is absent. After an ambiguous timeout, read the saved history; running/unknown work is never retried.

Both web and MCP use one account history. Counts come from native whole-result summaries. A+B is primary; A/B are diagnostics against the same input-eligible population, including matches. Samples can contain supporting and contradictory cases but never supply the denominator. One Binance USDT perpetual, six supported numeric fields; population also permits native daily price/OI crossings after its backend release. Outcome is +5% MFE within four hours, before costs/fills. Development is at most 31 days, separated by seven full days from reserved evaluation; anchors end by September 11, 2026. Reserved evaluation is never run.

The server retains definitions, failures, response bytes, HTTP/allowance headers and reproducibility keys. Default output projects examples and omits raw receipts, with recovery stated. Imported browser evidence stays labeled caller-supplied. Storage and concurrency errors stop further work; reopening never computes. Existing metering remains: API computations can each consume allowance, cached work is free; the session UI retains its existing population-scan meter.

Research contract

• Validation failures pass through as 422 {"errors":[{"code":"...","message":"..."}]}.
• Transport failures use the {"error","code"} envelope.
• Contract codes are machine-actionable. For errors such as UNSUPPORTED_FEATURE or OUTCOME_IN_PREDICATE, call list_features, repair the document, and retry.
• Deterministic tools are exact-document, UTC-only tools. interpret_prose may use a time zone to plan dates, but run_scan, run_cohort, and base_rate never reinterpret calendar language.
• Reruns and ETag 304 Not Modified revalidations are free. list_instruments ETags are scoped to the requested summary, symbol projection, or full representation.
• Interpretation is free and never debits the scan allowance. An unavailable scan allowance returns neutral 402 RESEARCH_ALLOWANCE_EXHAUSTED metadata without a checkout link.

REST API and documentation

The MCP server is a thin, deterministic interface to the public EdgeDepth Research API:

• REST API quickstart
• Authentication and API keys
• Credits, caching, and limits
• Versioned query grammar
• Reproducibility contract
• Worked API examples
• MCP connection guide
• How EdgeDepth Research works
• What you can ask

The default REST base used by the stdio package is https://app.edgedepth.com/api/v1/research.

Environment

Local stdio