Server has 4 tools with basic descriptions and partial schemas. All tools are READ_ONLY BigQuery operations. Naming is clear (verb_noun pattern: execute_, get_). Descriptions exist but are generic (avg ~120 chars, below 194-char baseline). Critical gaps: no input schemas visible in source code (only parameter names/types inferred from docstrings), no output schemas documented, no error recovery guidance, SQL injection vulnerability in get_client_details (string interpolation in WHERE clause), no parameter validation or constraints. Tool definitions appear to be inferred from decorators rather than explicit schema registration.

• Use parameterized queries (BigQuery client.query() with job_config.query_parameters) instead of f-string interpolation to prevent SQL injection in get_client_details and get_client_datasets.
• Document explicit JSON Schema for all tool inputs with constraints: execute_bigquery.query should have minLength/maxLength; client_id/client_name should have patterns or descriptions of valid formats.
• Add output schema documentation: specify that 'results' is a list of dicts with field names and types (e.g., 'results: List[Dict[str, Any]] with fields: client_id (str), client_name (str), sources (str), git_url (str)').
• Enhance error messages with recovery guidance: 'NotFound: Dataset not found. Verify dataset_id is correct and you have access. Try get_client_datasets() to list available datasets.' instead of bare 'Dataset or table not found'.
• Add pagination: accept limit (1-100, default 20) and offset parameters; return total_count and next_offset in response. Remove hardcoded LIMIT 1000.