Loading ToolBench
Meridian Edge MCP server — prediction market consensus data for AI assistants
Three tools with clear naming (get_consensus, get_opportunities, get_signals) and reasonable descriptions (150-200 chars each). Input schemas present with types and descriptions for all parameters. However, output schemas are completely undocumented, tools return formatted strings with no structured schema definition. Parameter descriptions lack constraint details (e.g., sport enum values stated in text but not formalized; limit ranges mentioned but not enforced). Error handling is present but generic. No tool annotations (readOnlyHint, etc.). Missing pagination guidance despite tools accepting 'limit' parameters.
Get real-time prediction market consensus probabilities. Returns aggregated consensus from multiple regulated prediction markets. Each event shows the collective probability, trend direction, and how much markets agree (spread).
Get events where prediction markets show notable divergence. Divergence opportunities are events where regulated prediction markets disagree significantly. Higher scores indicate greater disagreement. This may surface events where information is still being incorporated.
Get recent market signals showing direction of price moves. Signals indicate notable directional moves in prediction market consensus. Each signal shows whether the market shifted toward YES or NO, and the current status of the market.
Output schemas completely undocumented. All three tools return formatted strings with no structured schema definition. LLMs cannot parse results reliably or chain to downstream tools.
Parameter constraints stated in descriptions but not formalized. 'sport' accepts 'NBA, NFL, MLB, NHL, MLS, POLITICS' but no enum constraint in schema. 'limit' range '1 - 20' mentioned but not enforced via min/max.
No tool annotations. Tools are read-only (no state modification) but lack readOnlyHint annotation. Agents cannot infer safety properties without explicit hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Pagination not properly documented. Tools accept 'limit' but no mention of offset, cursor, or total_count in responses. Large result sets could exhaust context without clear pagination strategy.
Error handling returns plain text strings rather than structured error objects. LLMs cannot programmatically classify errors as retryable, user-fixable, or fatal.