A Model Context Protocol server for the Elsevier Scopus API
The Scopus MCP server registers 5 tools with complete JSON Schema input definitions and descriptions. Tool naming follows verb_noun patterns appropriately (search_scopus, get_abstract_details, get_author_profile, get_citing_papers, get_quota_status). However, significant gaps exist in parameter descriptions, output schema documentation, and error handling guidance. Most parameters lack detailed constraint descriptions that would guide LLM usage. Output schemas are completely undocumented, LLMs receive raw string representations of structured data with no declared field types or names. Error handling is minimal: exceptions are caught and converted to plain text error strings without recovery guidance or error classification. The server lacks input validation error messaging that would enable LLM self-correction (e.g., 'sort must be one of: coverDate, relevancy'). Parameter 'sort' is not declared as an enum despite accepting only specific values. No per-field documentation of response structure exists, forcing LLMs to infer what fields are available in returned data.
Retrieve full details for a specific document by Scopus ID.
Retrieve an author's profile by Author ID.
Retrieve a list of papers that have cited the specified document (Forward Citations).
Get the current API quota status (remaining/limit). Note: Values are updated only after making a request.
Search for documents in Scopus using a query string.
Output schemas completely undocumented. All tools return TextContent with raw string representations. LLMs receive no structured field definitions, forcing them to parse unstructured text. Pattern requires documenting return types with typed fields.
Parameter 'sort' in search_scopus and get_citing_papers accepts free-form strings without enum constraint. Description says 'e.g. coverDate, relevancy' but does not enumerate valid values. LLMs may hallucinate invalid sort orders.
Error messages lack recovery guidance. All exceptions return plain 'Error: {str(e)}' without classification (retryable vs user-fixable vs fatal) or actionable next steps. Pattern requires error responses to guide LLM recovery.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Parameter descriptions lack constraint details. 'count' parameter says 'default 5, max 25' but no minimum or explanation of what happens at limits. 'query' parameter in search_scopus lacks format/syntax guidance despite Scopus using domain-specific query language (TITLE, PUBYEAR, REFEID).
No input validation. Required parameters (query, scopus_id, author_id) are checked after extraction but error messages are unhelpful ('Query is required'). Missing format validation (e.g., valid Scopus ID format). Pattern requires detailed, actionable validation errors.
No pagination support documented. search_scopus and get_citing_papers return up to 25 results but no documentation of how to iterate through larger result sets, total counts, or next-cursor values. Large result sets could exceed token limits.
get_citing_papers silently modifies scopus_id by stripping 'SCOPUS_ID:' prefix and wrapping in REFEID() query. This transformation is undocumented. LLMs cannot predict this behavior, risking misuse if they pass unexpected ID formats.