Real-time market data server for crypto, equities, sentiment, and prediction markets
Market Pulse MCP has 6 tools with minimal parameter schemas and inconsistent descriptions. Most tools accept no parameters (empty input {}), making them inflexible. Descriptions are present but generic (10-50 chars), below the 50-200 char LLM-optimized baseline. No output schemas documented. No error handling guidance. Tool names follow verb_noun convention (get_*) but lack specificity about what data is returned. The server relies on external subprocess tools (yf, regime, polymarket) whose outputs are parsed with regex, fragile and undocumented. No input validation, no pagination support, no recovery guidance.
BTC, ETH prices with 24h change
SPY, QQQ prices with levels
Latest full market briefing as JSON
Daily ranges for BTC, ETH, SPY, QQQ
Current market regime (BULL/BEAR/CHAOS) + confidence
Top prediction markets by volume
No output schemas documented. LLMs cannot predict what fields to expect from any tool response, forcing them to parse unstructured output or make assumptions about downstream tool compatibility.
Descriptions are too short (10-50 chars) and lack context. 'Current market regime (BULL/BEAR/CHAOS) + confidence' does not explain WHEN to call this tool vs alternatives, what the confidence score means, or how to use the output. Baseline is 50-200 chars.
Most tools accept no parameters (empty input {}), making them inflexible. get_polymarket_top_markets accepts 'limit' but other tools cannot filter, paginate, or customize results. Agents cannot request specific symbols, date ranges, or result counts.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | <=2025-11-25 | v2 |
Regex-based output parsing (_parse_yf_price_output, _parse_regime_json) is fragile and undocumented. If upstream tool output format changes, parsing silently fails or returns incomplete data. No validation that parsed data matches expected schema.
No error handling or recovery guidance. If a subprocess times out (25s limit) or returns non-zero exit code, the tool raises RuntimeError with minimal context. LLM receives 'Tool timed out' with no suggestion to retry, check prerequisites, or use an alternative tool.