Loading ToolBench

Reading the current corpus…

superpath-demo MCP server · F · ToolBench

superpath-demo

MCP server for operational audit and ROI-ranked recommendations across sales, ops, and finance departments. Ingests CSV/JSON data, runs audits, and generates actionable recommendations.

2026-07-28+STDIO3 tools

44/100

Definition Quality49

Protocol Readiness50

Supportability33

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

Three tools with basic descriptions and minimal input schemas. load_sales_data has a single string parameter with description; run_audit and get_recommendations have empty input schemas ({}). Descriptions are present but brief (60-90 chars), lacking WHEN to use, prerequisites, or error recovery guidance. No output schemas documented. No parameter constraints (enums, ranges). Tool names follow verb_noun pattern (load_, run_, get_) but descriptions do not explain dependencies or state-mutation side effects. No tool annotations (readOnlyHint, destructiveHint). Error handling exists (isError flag) but messages are generic ('No data loaded'). Overall: functional but minimal, typical of early-stage demo servers.

Tools (3)

get_recommendationsread onlysource verified42/100

Turn the most recent audit findings into ranked, actionable recommendations for the sales org.

load_sales_dataread onlysource verified62/100

Ingest a client's sales data from a CSV or JSON file (columns: id, rep, stage, dealValue, daysInStage, outcome, createdAt).

run_auditread onlysource verified42/100

Analyze the currently loaded sales data for pipeline bottlenecks and low win-rate stages, ranked by estimated ROI.

Issues (5)

CRITICALtool ↗

run_audit and get_recommendations have empty input schemas ({}).

run_auditget_recommendations

HIGHresponse-shaper ↗

No output schemas documented. LLMs cannot infer what fields to expect from responses (e.g., structure of findings, recommendations). Blocks downstream tool chaining and forces LLM to parse unstructured JSON.

load_sales_datarun_auditget_recommendations

HIGHtool-description ↗

Descriptions lack WHEN to use and prerequisites. 'Analyze the currently loaded sales data' does not explain that run_audit requires load_sales_data to be called first, or that get_recommendations requires run_audit. Agents must infer dependencies.

run_auditget_recommendations

HIGHconstrained-input ↗

load_sales_data accepts filePath as a free-form string with no format constraints, validation rules, or examples. Description does not specify expected file format (CSV vs JSON), required columns, or error cases (file not found, malformed data).

Recommendations

Add explicit output schemas to all tools. Document the structure of findings (fields: id, stage, bottleneck_type, estimated_roi_usd, etc.) and recommendations (fields: rank, action, expected_impact, etc.). This enables LLMs to extract and chain results.
Expand descriptions to 100-150 chars and include dependency hints. E.g., run_audit: 'Analyze loaded sales data for bottlenecks and low win-rate stages, ranked by ROI. Requires load_sales_data to be called first.'
Add readOnlyHint: true to all three tools in the tool registration to explicitly declare they do not mutate state.
For load_sales_data, add format constraints: 'filePath must be a valid path to a CSV or JSON file. CSV columns: id, rep, stage, dealValue, daysInStage, outcome, createdAt. JSON: array of objects with same fields.'
Enhance error messages with recovery guidance. Instead of 'No data loaded', return 'No data loaded. Call load_sales_data(filePath) with a valid CSV or JSON file first.'

Score history

Overall score trend

First recorded score · v2 rubric

44/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	44	2026-07-28+	v2