Loading ToolBench

Reading the current corpus…

prompt-compass-mcp MCP server · C · ToolBench

← Browse

prompt-compass-mcp

github.com/rishabhmpc/prompt-compass-mcp

Prompt routing and firewall decisions from Prompt Compass over MCP.

2025-06-18+STDIO1 tools

69/100

Definition Quality77

Protocol Readiness50

SupportabilityNot measured-

Spec posture

Inferred effective spec 2025-06-18+·current baseline 2026-07-28

Static source inference · medium confidence · evidence: structured output

No deprecated protocol patterns detected

Summary

Single tool with clear naming (compass_check_prompt), explicit input schema with constraints (minLength 1, maxLength 262144), and documented output schema. Description is adequate (145 chars) but could be more LLM-optimized. Tool annotations present and accurate (readOnlyHint: false reflects quota metering). No parameter descriptions visible in schema, only the prompt parameter itself. Error handling delegates to API layer without recovery guidance. Test coverage demonstrates basic functionality.

Tools (1)

compass_check_promptread onlyauthsource verified77/100

Return a decision-only recommendation for whether prompt text should stay local, go to a cloud model, or be blocked for PII or jailbreak risk. This tool never sends the prompt to an AI model.

Issues (5)

HIGHtool-description ↗

Input parameter 'prompt' lacks description in schema. Only type and length constraints are visible; no guidance on what constitutes valid prompt text or when to call this tool vs alternatives.

compass_check_prompt

HIGHrecovery-guide ↗

Error responses from API are not documented. Tool description does not explain what errors are possible (e.g., invalid API key, quota exceeded, malformed prompt) or how the LLM should recover.

compass_check_prompt

MEDIUMtool-description ↗

Output schema fields lack descriptions. LLMs cannot infer meaning of 'decision' enum values (LOCAL_SAFE, CLOUD_COMPLEX, BLOCK_PII, JAILBREAK), 'confidence' range, or 'windows' semantics without explicit field documentation.

compass_check_prompt

MEDIUMtool-description ↗

Tool description does not clarify when to call compass_check_prompt vs other routing/firewall tools, or what downstream actions follow each decision type. Dependency hints are missing.

Recommendations

Add description to the 'prompt' parameter: 'The prompt text to check for routing and firewall decisions. Max 262KB. Supports any UTF-8 text.'
Document each output field: 'decision' (enum of routing/block decisions), 'blocked' (boolean), 'confidence' (0-1 score), 'quota' (remaining API calls), 'windows' (detection window count).
Add error recovery guidance: 'If the API returns 401, check PROMPT_COMPASS_API_KEY. If 429, quota is exhausted, wait before retrying. If 400, the prompt may be malformed or too long.'
Expand tool description to include decision semantics: 'LOCAL_SAFE: prompt is safe for local model. CLOUD_COMPLEX: route to cloud model. BLOCK_PII/JAILBREAK: do not forward prompt to any model.'
Migrate from STDIO to HTTP+SSE or Streamable HTTP transport to enable remote client integration and hosted deployment.

Score history

Overall score trend

First recorded score · v2 rubric

69/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	69	2025-06-18+	v2