Loading ToolBench

Reading the current corpus…

mcp-interactive-choice MCP server · C · ToolBench

← Browse

mcp-interactive-choice

github.com/oovz/mcp-interactive-choice

A MCP server for interactive user questions with native UI

2026-07-28+STDIO1 tools

60/100

Definition Quality82

Protocol Readiness50

Supportability30

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 'ask_user' with well-structured schema and clear descriptions. Tool naming follows verb-noun pattern (ask_user). Input schema is fully defined with Zod validation converted to JSON Schema. All parameters have descriptions. Output is documented via parseToolResult function. However, output schema is not explicitly documented in tool registration (only visible in implementation code), and tool lacks error handling guidance for recovery scenarios. No tool annotations (readOnlyHint/destructiveHint/idempotentHint) present.

Tools (1)

ask_userread onlysource verified82/100

Ask the user a question with several choices via a native GUI window. Supports Markdown in the body and a recommended choice. The user can also type a custom answer or skip the question entirely.

Issues (5)

HIGHtool ↗

Output schema not documented in tool registration. Tool definition shows input schema via Zod but parseToolResult implementation (returns string union of outcomes) is not declared to LLM.

ask_user

HIGHrecovery-guide ↗

Error handling does not guide LLM recovery. When 'recommended' doesn't match any choice, McpError is thrown but no alternative suggestions provided. Error message only states constraint violation, not corrective action.

ask_user

MEDIUMtool ↗

No tool annotations for state mutation. 'ask_user' modifies user-visible state (shows GUI window, solicits interaction) but lacks idempotentHint or any annotation clarifying retry safety and side effects.

ask_user

MEDIUMtool-description ↗

Parameter descriptions reference 'Markdown support' but do not specify format constraints (e.g., max length, allowed HTML tags, XSS handling). Code uses DOMPurify in UI but tool definition does not document sanitization guarantees.

ask_user

Recommendations

Add explicit output schema documentation to tool registration. Declare that ask_user returns one of: user-selected choice (string), custom_input (string), or cancellation/skip message (string). Include this in the tool's return_type or add a 'Returns' section to description.
Enhance error recovery guidance. When 'recommended' validation fails, return error message: 'Invalid recommendation: "%s" not in [%s]. Choose one of the listed options or omit this parameter.' This guides the LLM to self-correct.
Add tool annotations via MCP SDK. Use idempotentHint: false (calling twice shows window twice, different user answers) and document that this tool requires user interaction (not suitable for batch/background execution).
Document timeout behavior and constraints in description. E.g., '(Optional) GUI will close after timeout seconds if user does not respond. Timeout omitted means wait indefinitely.'
Add format constraints to 'body' parameter. E.g., '(Optional) Markdown text (max 5000 chars; script/iframe tags removed for security).' This clarifies sanitization to the LLM.
Specify 'choices' minItems and maxItems in schema. Current minItems=1 is good; consider adding maxItems (e.g., 100) to prevent GUI rendering issues with huge lists.

Score history

Overall score trend

↑ 8 points across a rubric change (v1 → v2)

60/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	C	60	2026-07-28+	v2
2026-03-09	D	52	-	v1