Public assessment

SidneyBissoli/medical-terminologies-mcp (medical-terminologies-mcp)

medical-terminologies-mcp · v1.5.7 · scanned Aug 3, 2026

What changed in the harness

Selection accuracy 98%, destructive-action safety rate 100% (baseline only -- no rewrite pass applied).

Category breakdown

Where the score comes from.

Earned points across the four signals Gradable measures. Safety and Legibility are scored out of 30; Economics and Discoverability are scored out of 20.

01Safety

30.0 / 30

30.0 out of 30

02Legibility

26.6 / 30

26.6 out of 30

03Economics

9.2 / 20

9.2 out of 20

04Discoverability

13.4 / 20

13.4 out of 20

Highest-impact fix

Estimated gain +11 points

Reduce the tool-definition and task token payload

Consolidate overlapping tools and trim repetitive schema text while preserving behavior, parameter, and return semantics.

Description evidence

Defects and rewrites.

0 defects found across the exposed tool descriptions. Suggested rewrites make purpose, inputs, boundaries, and returns easier for an agent to understand.

Tool	Defect types	Suggested rewrite
No description defects were flagged in this assessment.

Selection evidence

Confusable tool pairs.

14 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.

Tool A	Tool B	Confidence	Why they collide
`icd11_search`	`cid10_search`	high	The same disease term (e.g. diabetes) is valid input to both; a task like 'find the code for diabetes' is ambiguous between the international ICD-11 search and the Brazilian CID-10 search. cid10_search's own description warns readers to use icd11_search for the international revision, showing this confusion is real and expected.
`cid10_chapters`	`cid10_chapter`	medium	The names differ only by singular vs plural and both describe chapter navigation; a task like 'show me the CID-10 chapter information' is ambiguous between listing all 22 chapters and drilling into one chapter's constituent groups.
`icd11_chapters`	`cid10_chapters`	medium	A task like 'list the disease-classification chapters' does not specify ICD-11 vs Brazilian CID-10, and both tools return top-level chapter lists, so the agent can pick the wrong revision's chapter list.
`mesh_descriptor`	`mesh_tree`	medium	mesh_descriptor explicitly advertises 'view tree numbers showing hierarchy location', so a task like 'what is the tree number for D015242' can be routed to the descriptor tool instead of the dedicated mesh_tree tool, even though tree is the precise match.
`atc_lookup`	`atc_members`	medium	Both accept the same 1-5 character atc_code; a task like 'what is A10BA' is genuinely ambiguous between resolving the class name and level (lookup) and enumerating the member substances (members).
`icd11_search`	`icd11_lookup`	medium	A task like 'look up type 2 diabetes in ICD-11' supplies a disease name that only icd11_search can handle, but the verb 'look up' and the lookup tool's unpatterned code parameter can steer the agent to icd11_lookup; conversely 'get info on code BA00' may pull search.
`terminology_versions`	`terminology_diff`	medium	Tasks like 'what version information do you have for LOINC' do not clarify whether the user wants the current version and release metadata (versions) or the available version-to-version diff data (diff); the shared version theme makes either pick plausible.
`loinc_search`	`mesh_search`	medium	A bare query such as 'search for glucose' carries no system cue; both accept a free-text query, but mesh_search returns literature subject headings while loinc_search returns lab-test codes, so the wrong pick returns results from the wrong domain.
`icd11_lookup`	`icd11_hierarchy`	low	Both accept an ICD-11 code; a vague task like 'get the broader information and structure around BA00' is ambiguous between the full-definition tool and the parent/child navigation tool, though each description's focus is fairly distinct.
`loinc_details`	`loinc_panels`	low	Both take the same loinc_num; a task like 'tell me about the contents or structure of LOINC 24356-8' is ambiguous between the general details tool and the panel-structure tool, though the panel-specific wording should usually disambiguate.
`loinc_answers`	`loinc_panels`	low	Both return items associated under a LOINC code; a task like 'show me the options/items for this LOINC code' is ambiguous between questionnaire answer lists (answers) and panel components (panels), but the descriptions name distinct use cases.
`rxnorm_concept`	`rxnorm_ingredients`	low	Both take an rxcui; a task like 'what ingredients are in drug 161' is best served by rxnorm_ingredients, but rxnorm_concept advertises related ingredients and brands via include_related, so an agent seeking comprehensive detail may pick concept instead.
`atc_classify`	`atc_lookup`	low	Both descriptions contain 'look up'; a task like 'look up the ATC code for metformin' supplies a drug name that only atc_classify accepts, but the agent may misfire to atc_lookup, which expects an ATC code pattern the name does not satisfy.
`cid10_search`	`cid10_lookup`	low	A task like 'look up diabetes in CID-10' supplies a term that cid10_search handles and cid10_lookup cannot, yet the verb 'look up' maps onto cid10_lookup's name; ambiguity mostly arises when the user phrases a lookup-style request but supplies text instead of a code.

Compare the field

One score is useful.
The evidence makes it actionable.

Back to the leaderboard