Bybit Perpetual Futures Trading MCP Server
BybitTradingMCP has 4 tools with basic naming conventions and descriptions, but significant gaps in schema documentation, parameter descriptions, and output structure. Tools use action verbs (get_, analyze_) and descriptions are present, but output schemas are completely undocumented. Most parameters lack detailed descriptions of constraints, formats, or valid values. Error handling returns unstructured strings rather than actionable error guidance. The server reads API responses and formats them as plain text rather than structured JSON, limiting LLM reasoning capability. No tool annotations (readOnlyHint/destructiveHint) despite all tools being READ_ONLY. Overall quality is below the production baseline for agentic tools.
Performs a complete technical analysis of the market for a given trading pair.
Gets the current balance of the wallet.
Gets historical data (klines/candlesticks) for perpetual futures analysis.
Gets the current ticker (price, volume, etc) for a specific trading pair in perpetual futures.
Output schemas are completely undocumented. All tools return plain-text strings with no structured schema definition. LLMs cannot extract fields, plan downstream calls, or reason about response structure. This violates pattern:tool core requirement that LLMs know what to expect.
Parameter descriptions lack constraint details. 'intervalo' parameter accepts '1, 3, 5, 15, 30, 60, 120, 240, 360, 720, D, W, M' but description uses example syntax 'e.g.' instead of declaring an enum. This invites LLMs to hallucinate invalid values. 'limite' lacks min/max specification despite having min=200, max=1000 constraints mentioned only in docstring, not parameter schema.
Error handling returns unstructured error strings without recovery guidance. Examples: 'Error getting ticker: {retMsg}' or 'Error getting balance: {str(e)}'. Per pattern:recovery-guide, errors must tell the LLM what to do next. Current responses provide no actionable remediation (retry? call another tool? user action needed?).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Tool annotations missing. All 4 tools are READ_ONLY but lack readOnlyHint annotation. This prevents clients from optimizing caching, safety checks, and retry logic. Per spec, all tools should declare risk classification via annotations.
API credentials injected via environment variables but no indication in tool definitions or descriptions. While this follows pattern:secret-injection, the tool descriptions do not document that secrets are server-side managed, which could mislead users about security properties.
'analyze_market' description is vague: 'Performs a complete technical analysis of the market for a given trading pair.' Does not explain what 'complete' means, what indicators are calculated, what the output structure is, or when to use this vs get_klines. LLM cannot determine when to select this tool.