Upstream-read-only MCP server for NPLG Iverieli DSpace
Server has 8 tools with complete JSON schemas and basic descriptions. All tools are read-only or write operations with clear intent. Naming follows verb_noun convention (download_, get_, list_, search_, render_, inspect_). However, descriptions are uniformly terse (10-30 chars), lacking context on WHEN to use each tool, prerequisites, or return value structure. Parameter descriptions exist but are minimal. No output schemas documented. Error handling and recovery guidance absent. Composition is sound, tools are single-purpose and chainable via handle/render_id references.
Download a public PDF
Read full document metadata
Read a render manifest
Inspect a downloaded PDF
List document files
Create crop-only page tiles
Render PDF pages as JPEG
Tool descriptions are uniformly terse (10-30 chars). Baseline for A+ tools is 50-200 chars. Descriptions lack WHEN to use, prerequisites, and return value structure. LLMs cannot determine tool selection intent from 'Download a public PDF' alone.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what fields does get_document_metadata return? Is handle included for chaining?). Baseline: 100% of A+ tools document return types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Search NPLG Iverieli
render_pdf_pages and render_pdf_page_tiles accept 'pages' and 'tiles' array parameters with minimal description. No constraints on array size, tile format, or page number ranges. LLMs may pass invalid values (negative pages, malformed tiles) without guidance.
No error handling or recovery guidance. If search_documents returns no results, or download_document_file fails due to missing bitstream, LLMs have no actionable next steps. Baseline: error responses must tell LLM what to do next.
search_documents accepts optional 'cursor' parameter for pagination but no description of cursor format, total count, or next_cursor in response. Pagination contract is incomplete, risking context window exhaustion on large result sets.