The Pharmacology MCP server has 4 tools with clear naming following verb_noun convention (search_, get_), but suffers from significant structural and documentation gaps. All tools have basic descriptions (52-85 chars), meeting the minimum bar, but lack depth regarding WHEN to use them, WHAT they return, and HOW results should be consumed. Input schemas are present with types and descriptions for all parameters, but output schemas are entirely absent, critical for LLM chaining. All tools write to files as a side effect, but descriptions do not explicitly state this is an irreversible operation. Parameter descriptions are generic and lack constraints (enums, ranges, format specifications). No error handling guidance is visible. The server is usable for basic file-writing workflows but lacks the polish and completeness expected of production-grade agent tools.
Get all interactions for a specific ligand and save to a file
Get all interactions for a specific target and save to a file
Search for ligands/compounds and save results to a file
Search for pharmacological targets and save results to a file
No documented output schemas. Tools write results to files but LLMs have no visibility into what fields/structure the CSV or JSON contains. This breaks tool chaining, agents cannot extract and reuse results without extra parsing steps.
Descriptions lack actionable context. Phrases like 'Search for pharmacological targets and save results to a file' do not explain WHEN to call this vs similar tools, what the file format will be, or what happens on error. Descriptions should be 50-200 chars, optimized for LLM decision-making.
Destructive/side-effect operations (all tools write files) are not marked as such in descriptions. Descriptions must state 'modifies state' or 'writes to file' so agents know idempotency constraints and retry safety. No mention of what happens if the file already exists (overwrite? append? error?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 11 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 49 | - | v1 |
Parameter constraints are missing. 'target_type', 'ligand_type', 'species', 'interaction_type' accept free-form strings but should use enums. E.g., target_type should list allowed values (GPCR, Enzyme, Ion channel, etc.). Free-form strings invite hallucinated values.
No pagination or result limiting documented. Tools may return hundreds or thousands of results written to a file. No mention of how many results to expect, whether results are paginated, or what the size limit is. Agents have no way to know if a query will blow context.
No error handling guidance. Descriptions do not explain what to do if: file write fails, no results found, invalid ID passed, invalid filter values used. Error messages are not visible in the tool definitions.
Parameter descriptions lack format/constraint details. E.g., 'species' param says 'Filter by species (e.g., Human, Rat, Mouse)' but does not list all valid values or state it must be exact match vs case-insensitive partial match.
All tools require 'file_path_str' as output location but do not validate or document path constraints. No mention of: writable directory requirement, absolute vs relative paths, file format (CSV/JSON/TXT), or what happens if path is invalid/unwritable.