Four tools with basic schemas and descriptions, but significant gaps in parameter documentation and error handling. Tool names follow verb_noun convention (query, annotate_table, store, list_tables), which is good. However, parameter descriptions are minimal or missing entirely. The 'store' tool lacks clarity on what 'mode' values are valid beyond 'replace'. No output schemas documented. Error handling is absent, no guidance on what happens on SQL errors, constraint violations, or connection failures. Descriptions are present but generic (10-50 chars), below the 50-200 char LLM-optimized baseline. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY vs WRITE).

• Add enum constraints to 'mode' parameter in store tool: {"enum": ["replace", "insert", "upsert"]} with description 'Storage mode: replace (overwrite), insert (fail if exists), or upsert (insert or update).'
• Document output schemas for all tools. E.g., query returns {"rows": [{...}], "columns": [...], "rowCount": int}; store returns {"rowsAffected": int, "errors": [{"row": int, "message": string}]}.
• Expand tool descriptions to 80-150 chars. E.g., query: 'Execute a SQL query against the connected database. Returns rows and column metadata. Use for SELECT, aggregations, and read-only analysis.'
• Add tool annotations to schema: readOnlyHint=true for query/list_tables; destructiveHint=true for store; idempotentHint=true for annotate_table (idempotent overwrites).
• Add error handling guidance in descriptions. E.g., store: 'On constraint violation, returns per-row errors. On connection failure, returns retryable error with guidance to check FREEBASE_MCP_URL.'

Inferred effective spec: 2026-07-28+.

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