Loading ToolBench

Reading the current corpus…

tina_mcp_tools MCP server · F · ToolBench

tina_mcp_tools

A collection of MCP servers: DICOM Specialist (RAG agent for DICOM standard documentation) and UKSH Mensa (access to UKSH mensa meal plans)

2026-07-28+STDIO8 tools

49/100

Definition Quality62

Protocol Readiness42

Supportability33

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

This server has 8 tools across two domains (DICOM and mensa). Tool definitions are registered via @mcp.tool decorators with docstrings, but have significant quality gaps. All tools have descriptions and basic parameter typing, but parameter descriptions are minimal or absent. Output schemas are not formally documented. Error handling is not visible in the code, no recovery guidance, categorization, or actionable error messages. The tools follow basic verb_noun naming convention, but descriptions are inconsistent in depth. Parameters mostly lack detailed constraint documentation (formats, ranges, valid values). Composition is sound (each tool does one thing), but chaining IDs and references are not clearly documented in outputs.

Tools (8)

get_mensa_planread onlysource verified57/100

Get the UKSH mensa food plan for the specified date. Use tool "todays_date" first if you need todays date. To get other dates, use "offset_date" with todays date and the number of days to offset. e.g. if today is Monday, 2025/05/05, offset todays date by 4 days to get this Friday's date.

list_namesread onlysource verified67/100

Returns the names of all CIOD, modules or attributes

lookup_attributeread onlysource verified60/100

Find details for a Dicom attribute in a given CIOD domain

lookup_sop_nameread onlysource verified58/100

Returns the name for a given SOP ID

lookup_sop_uidread onlysource verified60/100

Search for the SOP UID by a given name

offset_dateread onlysource verified68/100

Offset the specified date by X days and / or weeks.

search_dicom_standardread onlysource verified63/100

Retrieve knowledge about the dicom standard by searching for a query.

Issues (7)

HIGHtool-schema ↗

Output schemas not formally documented. Tools return `list[str]`, `list[dict]`, `TimeResult`, or `str`, but no structured schema documentation for LLM reasoning about downstream usage. LLMs cannot reliably extract fields or plan chaining.

search_dicom_standardlookup_attributelookup_sop_namelookup_sop_uidlist_namestodays_dateoffset_dateget_mensa_plan

HIGHtool-description ↗

Parameter descriptions are minimal or missing. 'query' (search_dicom_standard), 'sop_uid' (lookup_sop_name), 'date' (offset_date, get_mensa_plan) lack context on format, valid values, or constraints. LLMs cannot infer what input format is expected.

search_dicom_standardlookup_attributelookup_sop_namelookup_sop_uidoffset_dateget_mensa_plan

HIGHrecovery-guide ↗

No error handling or recovery guidance visible. Code lacks try/catch blocks with actionable error messages. If a DICOM search fails or mensa PDF URL is not found, the LLM receives no direction on retry strategy or fallback.

Recommendations

Add formal output schema definitions. For each tool, document the return type as a JSON Schema (e.g., 'Returns: list of objects with fields {id: string, name: string, description: string}'). Use Pydantic BaseModel or explicit schema dictionaries in docstrings.
Expand parameter descriptions with format/constraint examples. E.g., 'query (str): Free-text search term (max 200 chars). E.g., "MRI imaging protocols"' and 'date (str): ISO 8601 date string, e.g., "2025-05-15"'.
Add error handling and recovery guidance. Wrap external calls (DuckDB queries, HTTP requests to mensa URL) in try/except blocks that return structured error responses with actionable next steps, e.g., 'DICOM search returned no results. Try: (1) broaden your search term, (2) call list_names to discover available attributes, (3) use lookup_sop_uid instead.'
Document valid values for enum-like parameters. For list_names.filter_by, list all valid values (at minimum: 'common', any CIOD name, or a wildcard pattern). Use Python Literal or explicit enum constraints.
Add pagination to list_names. Replace the hardcoded [:250] slice with a limit parameter (default=20, max=250) and return {results: [...], total: int, has_more: bool}.
Complete and document get_mensa_plan output. Show the full function signature and explain what the returned DataFrame/list contains (field names, types, example rows).
Simplify tool descriptions for LLM readability. Current descriptions are adequate length but could be more directive. E.g., 'lookup_attribute' could add: 'Use this to find DICOM attribute properties (VR, VM, retired status) within a specific domain. Call list_names(level="attribute") first to discover available attributes.'

Score history

Overall score trend

↑ 49 points across a rubric change (v1 → v2)

49/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	F	49	2026-07-28+	v2
2026-03-09	F	0	-	v1