Loading ToolBench

Reading the current corpus…

crowdcheck MCP server · C · ToolBench

crowdcheck

The contestation layer for last30days: turn engagement-ranked evidence into a consensus/confidence map an AI agent can act on.

2026-07-28+STDIO1 tools

61/100

Definition Quality81

Protocol Readiness50

Supportability35

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Single tool 'crowd_check' has a well-structured schema with all parameters typed and described. Tool name is clear and action-oriented. Description is comprehensive (280+ chars) and explains the tool's purpose, return value, and critical usage guidance. Schema includes proper enums for 'source' parameter and sensible defaults. However, output schema is not formally documented, the description mentions JSON fields but no structured schema is provided. Error handling is absent; no guidance on what happens if the claim is invalid, API fails, or demo mode is unavailable. The tool is read-only (good for safety) but lacks explicit error recovery patterns.

Tools (1)

crowd_checkread onlysource verified81/100

Map who agrees/disputes a claim and how sure the signal is. Args: claim: the assertion to check, e.g. "AI agents will replace most junior dev jobs by 2027". recency_days: evidence window in days (default 30). source: "auto" (default), "native" (Reddit+HN+market, no keys), or "last30days" for broader coverage (X, YouTube, TikTok) when that engine is installed. demo: use bundled offline sample evidence (no API keys needed). Returns: JSON contestation map: consensus, confidence, leaning, read, positions[], market_signal, coverage, evidence[]. NOT a verdict.

Issues (5)

HIGHtool ↗

Output schema not formally documented. Description mentions 'JSON contestation map: consensus, confidence, leaning, read, positions[], market_signal, coverage, evidence[]' but no structured schema definition is provided. LLMs cannot reliably parse or chain on undocumented output.

crowd_check

HIGHrecovery-guide ↗

No error handling or recovery guidance. Tool description does not explain what happens if the claim is invalid, API keys are missing, or the demo mode fails. LLMs have no guidance on retryability or next steps on failure.

crowd_check

MEDIUMtool-description ↗

Parameter 'source' enum includes 'last30days' which requires an external engine to be installed, but this dependency is not validated or documented in the tool. LLMs may select this option without knowing it will fail if the engine is absent.

crowd_check

MEDIUMresponse-shaper ↗

Tool returns JSON as a string (via json.dumps) rather than a structured object. This forces LLMs to parse JSON text, increasing error likelihood and token waste. Return structured output directly.

Recommendations

Document the output schema formally. Define the structure of ContestationMap with field types, e.g. {consensus: float [0-1], confidence: float [0-1], positions: [{stance: string, count: int, engagement: float}], market_signal: {odds: float, source: string}, evidence: [{text: string, platform: string, engagement: int, date: string}]}.
Add error handling guidance to the tool description: 'Returns an error if the claim is empty, if the requested source engine is not installed, or if the API is unreachable. Retry with demo=true for offline testing.'
Validate the 'source' parameter at runtime. If 'last30days' is selected but the engine is not installed, return a clear error: 'Engine last30days not installed. Use source="native" or source="demo" instead.'
Return structured output (dict/object) instead of JSON string. Change the tool to return the ContestationMap object directly; let the MCP framework serialize it.
Migrate from STDIO to HTTP or HTTP+SSE transport to enable remote accessibility and compatibility with hosted MCP clients.

Score history

Overall score trend

First recorded score · v2 rubric

61/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	61	2026-07-28+	v2