Loading ToolBench
Smart-money market structure (ChoCh/BOS, liquidity sweeps, FVG/OB, pullback zones) as an MCP server. The same engine that runs in production at TradeFloor.
Three tools with complete input schemas and reasonable descriptions (120-180 chars each). All parameters have types and descriptions. However, output schemas are not formally documented in the tool definitions, only visible in the implementation code. Tool names follow verb_noun pattern (analyze_, find_) but lack specificity about what structure/blocks/sweeps means to a non-domain expert. Parameter descriptions are domain-specific but could better explain when/why to use each tool. No error handling guidance visible in tool definitions.
Read market structure: trend, character shifts (ChoCh) and continuation breaks (BOS). Returns the current trend, every structural break in the window with its time and price, the last validated swing high and low, the potential swing levels a pullback is forming against, and the impulse peak or trough of the current leg. Give a coin (BTC, ETH, SOL; tokenised equities as xyz:AAPL) to fetch candles from Hyperliquid's public endpoint, or pass your own candles, or neither to see a worked example on bundled ETH-15m data. Set include_candles=true when you intend to DRAW the result: the response then carries the candles themselves, on the same time scale as the levels.
Find fair value gaps, order blocks and the pullback zones holding them. Reports blocks on the current impulse leg, then again per Fibonacci pullback zone (discount, equilibrium, premium, golden pocket), and says which zones price is sitting in right now. Same input options as analyze_structure.
Find stop-hunts: breaks that first ran the liquidity resting past a level. A sweep is a structural break whose candle pushed through one or more prior swing levels before closing back inside. Each result names the sweeping candle and every level it took out, which is what separates an engineered break from a clean one. Same input options as analyze_structure.
Output schemas not formally documented in tool definitions. Only visible in implementation code (tools.py return dicts). LLMs cannot plan downstream tool calls or extract fields without explicit schema documentation.
Tool names use domain jargon (ChoCh, BOS, FVG, OB) without explanation in descriptions. LLMs unfamiliar with SMC trading concepts cannot determine when to select these tools vs alternatives.
No error handling guidance in tool definitions. If Hyperliquid API fails, candle parsing fails, or invalid coin symbol is passed, LLMs have no recovery path documented.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | <=2025-11-25 | v2 |
Parameter 'candles' accepts array of objects but does not specify required fields (open, high, low, close, time). LLMs may pass malformed candle data.
Tool descriptions mention 'worked example on bundled ETH-15m data' but do not explain how to trigger it or what the response looks like. Discovery is unclear.