Loading ToolBench
Crypto intelligence for AI agents — exposes signal triggers, factor scores, market analysis, and backtesting tools via the Model Context Protocol. All tools proxy to the CRYPTYX REST API.
CRYPTYX MCP Server demonstrates solid fundamentals with 13 well-named tools, consistent schema declarations using Zod, and comprehensive parameter descriptions. All tools follow verb_noun naming conventions (get_*, search_*) and include domain-specific parameter constraints (enums, ranges). However, output schemas are completely undocumented, responses are passed through textResult() without any documentation of what fields clients should expect. Error handling is minimal (raw API error messages with no recovery guidance). Missing tool annotations (readOnlyHint/destructiveHint), no documented output structure, and responses lack pagination/result limiting despite tool descriptions suggesting potentially large datasets (e.g., 'all 35 signals', '~200 assets').
Single-metric z-score backtest with forward returns across 8 horizons (1d to 365d). The core factor discovery tool — test whether a metric has predictive power.
Run a backtest for a signal over a date range. Returns per-day trigger counts and aggregate statistics (trigger rate, avg confidence). Essential for strategy optimization.
Get the full current-state snapshot optimized for AI agents: factor breadth, top/bottom composite rankings, signal trigger summary, and pipeline status.
Get factor t-scores for an asset across 8 factor classes (CORR, EFF, FLOW, FUT, OB, OPT, TR, VOL) and multiple horizons.
Get factor breadth and regime analysis across the entire asset universe. Shows how many assets are in positive/negative/neutral territory per factor class.
No output schemas documented for any tool. Responses are JSON.stringify() with no specification of field structure, types, or nested object definitions. Clients cannot validate or plan downstream calls based on response shape.
Error handling is minimal. Raw API errors ('CRYPTYX API 404: ...') are surfaced with no recovery guidance. Errors do not indicate whether they are retryable, user-fixable, or fatal. No suggestions for alternative tools or corrective actions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Get asset universe with composite scores, returns, and rankings. Returns top-ranked digital assets by CRYPTYX composite score.
Get daily OHLCV candle data for a single asset. Useful for backtesting and charting.
Get the current regime classification for an asset (e.g. trending, mean-reverting, volatile). Includes primary and secondary regime with confidence scores.
List all 35 signals with their active parameters and 30-day trigger statistics. Useful for understanding available signals before backtesting.
Get a structured explanation of why a specific signal fired (or did not fire) for a given asset on a given day. Returns factor scores and composite context.
Get today's active signal firings across all assets. Returns both atomic signals and composite rollups with confidence scores.
Scan a metric across all ~200 assets for z-score extremes on the latest day. Returns ranked results with forward returns at 1d/7d/30d horizons.
List all tracked assets in the CRYPTYX universe with their universe tags.
No result limits or pagination documented. Tools like scan_metric_universe ('~200 assets') and get_market_snapshot (all assets) likely return large result sets without documented limit, offset, or cursor parameters. This risks exhausting context windows.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). backtest_signal modifies state ('write to signal_log') but is not marked destructive. Clients cannot distinguish safe vs. risky operations. All 13 tools appear read-only in descriptions, but this is not enforced via annotations.
backtest_signal returns per-day trigger counts but aggregate statistics are not documented. Description promises 'trigger rate, avg confidence' but response structure is undefined.
STDIO transport only. Server is not remotely accessible and cannot be used by hosted or managed MCP clients. Protocol readiness capped at 50 per HARD SCORING RULES.