AlphaFold DB MCP has basic tool definitions with schemas and descriptions, but significant gaps limit production readiness. All 4 tools have input schemas with type definitions and descriptions present. However, naming lacks action verbs (prediction, summary, annotations, uniprot are nouns/identifiers, not verb_noun patterns like get_prediction). Parameter descriptions are minimal (10-30 chars, well below the 72-char baseline). Output schemas are completely undocumented, LLMs cannot plan downstream calls or extract structured data. Error handling is generic (404 → 'not found', non-200 → status code + truncated text). No parameter validation guidance, no enum constraints, no pagination support despite API returning potentially large result sets. The 'qualifier' parameter appears in 3 tools but lacks format guidance (is it case-sensitive? What's the valid format?). Tools 'prediction' and 'uniprot' appear to be duplicates with different parameter names, violating single-responsibility and composition patterns.
Sequence-level annotations (e.g. MUTAGEN, LOW_CONFIDENCE_REGIONS).
Full prediction metadata + structure file URLs (PDB, CIF, PAE).
Short summary for a prediction (organism, sequence, mean pLDDT).
Fetch AlphaFold structure prediction metadata (PDB/CIF/PAE download URLs, pLDDT scores) by UniProt accession (e.g. "P00533"); equivalent to `prediction` but takes uniprot_id explicitly.
Tool names lack action verbs. 'prediction', 'summary', 'annotations', 'uniprot' are nouns/identifiers, not verb_noun patterns. LLMs infer intent from names, 'get_prediction', 'get_summary', 'list_annotations', 'fetch_uniprot_structure' would be clearer.
Output schemas completely undocumented. No description of what fields are returned, their types, or structure. LLMs cannot plan multi-step workflows or extract required IDs for chaining.
Parameter descriptions are minimal (10-30 chars vs 72-char baseline). 'qualifier' in 3 tools lacks format guidance, is it case-sensitive? Must be exact UniProt accession? No constraints or examples.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Duplicate tools: 'prediction' and 'uniprot' appear functionally identical (both fetch AlphaFold metadata by accession), just with different parameter names ('qualifier' vs 'uniprot_id'). Violates single-responsibility and composition patterns.
Error handling is generic. 404 → 'not found', non-200 → status + truncated text. No guidance on what to do next (retry? ask user? check accession format?). No error classification (retryable vs fatal).