Loading ToolBench

Reading the current corpus…

mcp-reference MCP server · C · ToolBench

mcp-reference

MCP server for extracting, analyzing, and managing references and citations in Markdown documents. Provides tools for reference extraction, citation detection, index generation, and recommendation.

<=2025-11-25STREAMABLE_HTTP6 tools

63/100

Definition Quality64

Protocol Readiness62

SupportabilityNot measured-

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging

Deprecated protocol patterns detected

Summary

Server has 6 tools with basic FastAPI/FastMCP integration. Tool names lack action verbs (health, stop, extract_references, extract_citations, index_generate, recommendations). Descriptions exist but are in Chinese and vary in quality (10-200+ chars). Input schemas are present with type definitions, but parameter descriptions are minimal or missing. No output schemas documented. Error handling is basic (HTTPException for missing params). Security concern: stop() tool allows unauthenticated process termination. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Composition is reasonable (single responsibility per tool), but naming conventions deviate from verb_noun pattern.

Tools (6)

extract_citationsread onlysource verified68/100

从 Markdown 内容或文件路径中提取所有引用位置与类型，并生成与参考文献的映射上下文

extract_referencesread onlysource verified72/100

从 Markdown 内容或文件路径中提取参考文献项，可选择按 DOI、标题或作者-年份进行合并去重，并返回解析结果与样式统计

healthread onlysource verified60/100

检查服务是否正常运行

index_generateread onlysource verified63/100

按首次引用顺序重新编号，并返回编号映射与新的参考文献列表

recommendationsread onlysource verified68/100

基于 Markdown 内容与已解析参考文献，给出 Top-K 相关文献推荐

stopdestructivesource verified53/100

触发服务停止并退出进程

Issues (5)

HIGHtool-naming ↗

Tool names lack action verbs. 'health', 'stop', 'recommendations' do not follow verb_noun pattern. Should be 'check_health', 'shutdown_server', 'get_recommendations'.

healthstoprecommendations

CRITICALpermission-gate ↗

Destructive tool 'stop' has no authentication, permission gate, or confirmation step. Any caller can terminate the server process.

stop

HIGHtool-annotations ↗

No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels declared in metadata. Tools should declare risk via schema annotations.

healthstopextract_referencesextract_citationsindex_generaterecommendations

HIGHresponse-shaper ↗

Output schemas not documented. Responses are Pydantic models (ExtractReferencesResponse, etc.) but no schema definitions visible in source. LLMs cannot infer response structure.

Recommendations

Rename tools to follow verb_noun pattern: 'check_health', 'shutdown_server', 'get_recommendations', 'list_references' (if applicable).
Add permission gate to 'stop' tool: require admin role or API key. Implement dry-run confirmation pattern before process termination.
Document all output schemas explicitly. Show field names, types, and descriptions for ExtractReferencesResponse, ExtractCitationsResponse, etc.
Expand parameter descriptions with format/constraint details. E.g., 'merge_by: Strategy for deduplication. doi=match by DOI field; title=exact title match; author_year=first author + publication year.'
Add tool annotations to schema: mark 'stop' with destructiveHint=true, 'extract_*' with readOnlyHint=true, and 'index_generate' with idempotentHint=true.

Spec posture evidence

Inferred effective spec: <=2025-11-25.

Relies on Logging (deprecated) - log to stderr or use OpenTelemetry

Score history

Overall score trend

First recorded score · v2 rubric

63/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	63	<=2025-11-25	v2