MCP server for ProFunding — AI-powered funding rate arbitrage across 20+ perp DEXes
ProFunding has 5 tools with clear verb-noun naming (get_*, run_backtest). All tools have descriptions (10-100 chars, baseline 194 avg) and input schemas with types. However, descriptions are terse (e.g., 'Get live funding rate arbitrage opportunities across all DEXes. [Free]', 73 chars), lacking WHEN/WHY context. Parameter descriptions are present but minimal (e.g., 'Minimum net APR to filter (default 0)', 39 chars, baseline 72 avg). No output schemas documented. Error handling is absent, no recovery guidance, no categorization. Tools return formatted strings rather than structured objects, forcing LLM parsing. No tool annotations (readOnlyHint, destructiveHint). Composition is sound (each tool does one thing), but output design wastes tokens on human-readable formatting instead of structured JSON.
List all connected DEXes with their status and funding interval. [Free]
Get historical funding rates for a symbol on a specific exchange. [Free]
Get live funding rate arbitrage opportunities across all DEXes. [Free]
Get funding rate spread chart data for a pair across two exchanges. [Free]
Backtest a delta-neutral funding arbitrage trade with historical data. [Free]
No output schemas documented. Tools return formatted strings (e.g., 'Found 3 opportunities:\n ETH/USDC, Long Hyperliquid...') instead of structured JSON. LLMs must parse unstructured text, wasting tokens and increasing hallucination risk.
Descriptions are terse (40 - 73 chars) and lack WHEN/WHY context. E.g., 'List all connected DEXes with their status and funding interval. [Free]' does not explain when to call this vs get_opportunities. Baseline is 194 chars.
Parameter descriptions are minimal (e.g., 'Minimum net APR to filter (default 0)', 39 chars). Missing format/range guidance. E.g., 'days' param lacks bounds (1 - 90 stated in get_historical_rates but not in run_backtest). Baseline is 72 chars per param.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If API returns 'No opportunities found', the tool returns a plain string. No guidance on what to try next (e.g., 'Try lowering min_apr or removing filters'). No categorization of retryable vs user-fixable errors.
No tool annotations. All tools are read-only (per Risk field), but readOnlyHint is not declared in schema. This forces LLMs to infer safety from descriptions, increasing misuse risk.