The OpenCaseLaw MCP server exposes 45 tools across legal research domains (decisions, legislation, scholarship, materials). However, the evaluation reveals significant quality gaps across naming, descriptions, parameter validation, output schemas, and error handling. While tool names generally follow verb_noun conventions (search_decisions, get_law, find_citations), the overwhelming majority lack substantive descriptions and parameter-level documentation. The source code shows tools are listed in scripts/tool_surface_check.py with minimal inline documentation. Many tools expose generic parameters (query, limit) without type constraints, enums, or validation guidance. No output schemas are documented in the visible code, and error handling patterns are absent. The server appears to be a research/academic prototype with good domain coverage but production-grade quality gaps.

• Rename 'search' to 'search_all_sources' and 'fetch' to 'fetch_document_by_id' to disambiguate intent.
• Add full output schema documentation for all 45 tools. Include field names, types, example values, and pagination metadata.
• Expand all parameter descriptions to 50-150 characters with format constraints, valid value enums, and default behavior.
• Add 'limit' parameter guidance: specify default (e.g., 20), maximum (e.g., 100), and pagination mechanism (cursor/offset/page).
• Document error scenarios for each tool: 'If query is empty, returns error code 400 with message: "query parameter required". If no results found, returns empty array (not an error).'
• For tools accepting 'language' parameter, use JSON Schema enum constraint instead of prose example: {"enum": ["de", "fr", "it"]}.
• For 'fields' parameter in search_decisions, replace vague 'compact' example with explicit enum: {"enum": ["compact", "full", "metadata_only"]} with descriptions of each.
• Add idempotency guarantees: 'Calling fetch with the same ID repeatedly returns identical results (safe to retry).'
• Document rate limits, if any: 'Agents may call search_* tools up to 100 times per minute. Exceeding this returns 429 Too Many Requests with Retry-After header.'
• For generative tools (draft_mock_decision, generate_exam_question), add mandatory disclaimers: 'Output is synthetic and for educational/illustrative use only. Not a substitute for professional legal advice. May contain factual errors.'
• Add chaining guidance: 'search_decisions returns decision IDs in result.decision_id. Pass to get_decision(decision_id) to retrieve full details.'

Inferred effective spec: 2026-07-28+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained