A legal assistant that can answer questions about your documents and more.
Server has 6 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-100 chars, mostly adequate) and Zod-based input schemas with type definitions. However, output schemas are undocumented, responses return unstructured text or raw API responses without field documentation. Error handling is present but generic ('Error en el proceso'). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Parameter descriptions are minimal (1-2 sentences). No pagination support for list tools. Tool naming uses prefixes (lar-*) rather than verb_noun pattern, reducing clarity. Composition is reasonable, tools are single-purpose, but output chaining is weak (no IDs returned to enable downstream calls).
Tool to answer questions about your documents using Legal Assistant RAG.
Tool to create a new client in the Legal Assistant RAG system.
Tool to delete a document from the Legal Assistant RAG system.
Tool to list all available clients in Legal Assistant RAG.
Tool to list all available documents in Legal Assistant RAG.
Tool to upload a document from a URL to Legal Assistant RAG.
Tool names do not follow verb_noun convention. All tools use 'lar-*' prefix (lar-ask, lar-upload-document, lar-delete-document) instead of action verbs (ask_document, upload_document, delete_document). This reduces LLM clarity on intent and violates baseline naming patterns (90% of A+ tools start with action verb).
Output schemas are completely undocumented. Tools return unstructured text in 'content[0].text' or raw API responses (e.g., apiResponse field in lar-upload-document). LLMs cannot plan downstream calls or extract structured data. No field documentation for return values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
No tool annotations despite clear risk levels. lar-delete-document is DESTRUCTIVE but lacks destructiveHint. lar-ask and lar-list-* are READ_ONLY but lack readOnlyHint. lar-upload-document and lar-create-client are WRITE but lack idempotentHint documentation. Agents cannot determine retry safety.
List tools (lar-list-documents, lar-list-clients) have no pagination support. No limit, offset, or cursor parameters. Descriptions do not mention result limits. If 1000+ documents exist, response will be massive and blow context window.
Error messages are generic and non-actionable. Most errors return 'Error en el proceso: {error.message}' or HTTP status + raw text. No guidance on retry, user action, or recovery. Does not follow error-classification pattern.