Loading ToolBench

Reading the current corpus…

mcp-app-debug MCP server · F · ToolBench

mcp-app-debug

Debug both ends of MCP Apps — render your server's app in a Playwright browser with 7 automated diagnostics, 12 under --profile (npx mcp-app-debug <server-url>), or grade a HOST's MCP Apps conformance with a probe fixture server (npx mcp-app-debug host). Speaks the 2025-11-25 and stateless 2026-07-28 MCP revisions, and --aggregator reproduces the namespacing-gateway failure that breaks apps calling their tools by the bare name.

2026-07-28+STREAMABLE_HTTP4 tools

39/100

Definition Quality22

Protocol Readiness75

Supportability43

Spec posture

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

Static source inference · medium confidence · evidence: stateless requests, Apps extension, Streamable HTTP

Current-spec patterns detected

Summary

This is a debug/diagnostic tool, not a production agent tool server. The 4 tools have minimal descriptions (10-50 chars), no parameter descriptions, and schemas are either empty objects or single undescribed properties. Tool names are generic (probe, report, poke, forecast) without action verbs. The 'report' tool is explicitly marked 'Not for the model', it's an internal callback, not an agent-facing tool. No error handling guidance, no output schemas documented. This server is designed for MCP protocol conformance testing, not for LLM agent interaction.

Tools (4)

forecastread onlysource verified38/100

Returns a forecast for a city.

pokeread onlysource verified17/100

Returns a poke (or an error, in tool-error scenario).

proberead onlysource verified15/100

Renders the mcp-app-debug probe app, which grades this host's MCP Apps support. Call it with no arguments.

reportread onlysource verified18/100

Called by the probe app to report what it observed. Not for the model.

Issues (5)

CRITICALtool ↗

probe and poke tools have empty input schemas ({}).

probepoke

CRITICALtool-description ↗

No parameter descriptions anywhere. 'forecast' accepts 'city' with a default but no description of what city means, format, or constraints.

forecast

HIGHtool-description ↗

Tool descriptions are under 50 characters and lack context. 'probe' says 'Renders the mcp-app-debug probe app' but does not explain when to call it or what it returns. 'report' is explicitly marked 'Not for the model', should not be exposed as an agent tool.

probereport

HIGHtool ↗

Tool names lack action verbs. 'probe', 'report', 'poke', 'forecast' do not follow verb_noun convention (e.g., run_probe, submit_report, test_poke, get_forecast). LLMs cannot infer intent from these names.

probereportpokeforecast

Recommendations

Add parameter descriptions to 'forecast': 'city (string, default Tokyo): The city name for which to retrieve weather forecast. Accepts city names or ISO 3166-1 alpha-2 country codes.'
Rename tools to follow verb_noun convention: 'probe' → 'run_probe', 'report' → 'submit_report', 'poke' → 'test_poke', 'forecast' → 'get_forecast'.
Expand tool descriptions to 100-200 chars explaining purpose, when to call, and what is returned. E.g., 'probe': 'Runs the MCP Apps conformance diagnostic suite. Returns structured results grading the host's MCP Apps support. Call this to validate host implementation.'
Document output schemas for all tools. At minimum, specify the top-level type (object/array) and key fields returned.
Remove 'report' from agent-facing tools or clearly mark it as internal-only in the server's capabilities. It is a callback channel, not a user-invoked tool.

Spec posture evidence

Inferred effective spec: 2026-07-28+.

Current-spec patterns present: stateless request handling, structured tool output, Streamable HTTP

Score history

Overall score trend

First recorded score · v2 rubric

39/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	39	2026-07-28+	v2