Loading ToolBench
AI-powered crypto trading system with pluggable signal fusion and MCP server integration for Binance, macro, on-chain, and news data
This server provides 10 tools for cryptocurrency market analysis. While tool naming follows verb-noun conventions (get_*, search_, load_), there are critical gaps in schema completeness and parameter documentation. Most tools lack visible input schema validation in the provided source code. Descriptions are present but many are verbose and include example values (anti-pattern). Parameters like 'provider_key' across multiple tools are poorly documented, users won't know what API provider to specify or how to obtain credentials. The 'load_skill' tool is poorly defined and lacks clarity on available skill names. Schema documentation for outputs is entirely absent, LLMs cannot plan downstream calls without knowing response structure. Error handling is minimal; no guidance on recovery actions or retryability. Overall, this is a D-grade server suitable for local experimentation but not production agent deployment.
Fetch total DeFi TVL and 7-day change from DefiLlama. Use to assess capital flows in the DeFi ecosystem.
Fetch real-time derivatives data from Binance (OI, long/short ratio, taker ratio). Use when you need current futures positioning data.
Fetch exchange net inflow/outflow data. Positive = net inflow (sell pressure), Negative = net outflow (accumulation).
Fetch the current Crypto Fear & Greed Index (0-100). Extreme Fear (<25) is a contrarian buy signal. Extreme Greed (>75) is a contrarian sell signal.
Fetch recent funding rate snapshots from Binance perp futures via ccxt. Use to detect persistent funding bias (sustained positive = crowded long).
Fetch 24h liquidation data (long vs short liquidation amounts). Use to assess leverage flush risk and cascade potential.
Missing output schema documentation for all tools. LLMs cannot plan chained calls without knowing response structure (fields, types, pagination). Every tool lacks documented return types.
Opaque 'provider_key' parameter (liquidation_data, whale_transfers, exchange_netflow) lacks guidance on which provider to use, how to obtain credentials, or whether it's required. This forces agents to guess or fail at runtime.
Descriptions include example values ('BTC/USDT', '8h periods', 'CoinDesk, CoinTelegraph') which LLMs may reuse literally rather than adapting to context.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Fetch social buzz metrics from CoinGecko (Twitter followers, Reddit, sentiment). Use to gauge community engagement and sentiment momentum.
Fetch recent large whale transfers for a given asset. Use to detect smart money accumulation or distribution.
Load the body content of a named skill. Returns the skill body as a string, or an error message.
Search recent crypto news articles from CoinDesk, CoinTelegraph, Decrypt. Returns articles with title, summary, source, and key events.
load_skill tool is poorly defined. Description does not explain what skills exist, how to discover available skills, or what the response format is. Parameter 'name' lacks enum constraint or discovery mechanism.
Parameter descriptions are generic or missing constraints. E.g., 'pair' accepts 'like BTC/USDT' but does not state which exchange notation is expected, whether lowercase is valid, or what happens if an unsupported pair is passed.
No error handling or recovery guidance in any tool description. LLMs receive no instructions on what to do if a tool call fails, whether to retry, or how to self-correct.
Missing pagination documentation for search_crypto_news. No indication of total result count, next_cursor, or behavior when limit is exceeded. Agent cannot safely loop through results.
No tool composition or chaining guidance. It is unclear which tool outputs feed into which tool inputs, forcing agents to reason about data flow without help.