Single-tool server with strong naming, comprehensive descriptions, and proper schema definition. Tool name 'get_neo4j_schema_and_indexes' follows verb-noun convention and clearly signals discovery purpose. Description is detailed (280+ chars) and includes actionable guidance about calling before search tools, property warnings, and token limits, well above median (194 chars). Parameter 'sample_size' has type definition and description. ToolAnnotations properly set (readOnlyHint=true, destructiveHint=false, idempotentHint=true, openWorldHint=true). However, no output schema is explicitly documented in the code snippet, only implied through the query logic (vector indexes, fulltext indexes, schema metadata, property size warnings). The implementation infers structure from Neo4j APOC calls but does not formally document the return type or field names for downstream tool composition. No error recovery guidance in the description (e.g., what to do if APOC is unavailable). Missing dependency documentation about Neo4j versions requiring APOC.

• Document the full return schema. Add a Python TypedDict or Pydantic model showing the exact structure: e.g., {"vector_indexes": [...], "fulltext_indexes": [...], "node_schema": {...}, "relationship_schema": {...}, "property_warnings": {...}}. Include field names, types, and examples so LLMs know what to extract.
• Add error recovery guidance to the description. Example: 'If APOC is not installed, the query will fail with a Neo4j error. Ensure the apoc library is loaded. If no indexes are found, any results are empty, call vector_search or fulltext_search tools anyway; they may still work if indexes are not explicitly shown.'
• Add APOC version requirement to the description. Example: 'Requires Neo4j APOC library (5.0+). If you see 'Unknown procedure: apoc.meta.schema', install APOC on the Neo4j instance.'
• Consider a 'return_properties' parameter to allow filtering which fields to return (e.g., only vector indexes, no property warnings) to reduce token overhead when not needed.
• Document what sample_size=0 means. If it defaults to config_sample_size, clarify: 'Defaults to 1000. Larger values (e.g., 5000) are more accurate but slower; smaller values (e.g., 100) are faster but may miss properties on rare nodes.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry