This server presents a strong set of 17 well-named database and Neon management tools with comprehensive input schemas and mostly good descriptions. However, several critical gaps prevent a higher score: (1) Tool #14 (inspect_database) has NO visible input schema in the provided code, (2) Tool #17 (fetch) accepts a bare 'url' parameter with minimal context about valid formats, (3) Several tools lack explicit output schema documentation, (4) Error handling patterns are not evident in the code samples, and (5) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are visible despite the toolkit supporting them. The naming is consistently strong (verb-noun patterns like run_sql, describe_table_schema, prepare_database_migration). Descriptions are generally 50-150 characters, good length for LLM selection. Most parameters have types and descriptions, but some optional fields lack guidance on when to use them (e.g., role_name in get_connection_string). Security patterns are present (scope-based filtering via grant-filter.ts, tool annotations framework), but not fully leveraged in the visible implementation.

• Add input schema for inspect_database. Specify which checks to run (e.g., 'table_count', 'index_usage', 'constraint_violations') as an enum or string array, and document the expected output structure.
• Annotate all tools with readOnlyHint, destructiveHint, and idempotentHint metadata. For example: run_sql_transaction should have destructiveHint=true; run_sql with readOnly=true should have readOnlyHint=true.
• Expand fetch description to 80-150 characters. Document supported schemes (http, https), rate limits (e.g., 'max 10 requests/minute'), content-type handling, and max response size. Example: 'Fetch documentation content from URLs. Supports http(s) only. Max 1MB response. Rate limited to 10 req/min.'
• Add explicit output schema to each tool definition. Use JSON Schema to describe field names, types, and meanings. Example for run_sql: {type: 'object', properties: {rows: {type: 'array'}, rowCount: {type: 'number'}, fields: {type: 'array'}}}.
• For complete_database_migration and complete_query_tuning, document parameter dependencies in a 'Parameters' section. Example: 'temporary_branch_id and parent_branch_id are required together; apply_changes controls whether to commit.'
• Add error guidance to tool descriptions. Example for run_sql: 'If the query fails, a specific SQL error is returned. Common issues: table not found, call get_database_tables first; permission denied, check your role via describe_branch.'
• Provide pagination guidance for list_slow_queries and list_organizations. Explicitly state: 'Returns up to `limit` items (default 10). Use offset for next pages.'

Inferred effective spec: 2026-07-28+.

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