MCP (Model Context Protocol) server for MariaDB and MySQL databases. Enables Claude Code and other MCP clients to interact with databases through a secure, standardized interface.
MariaDB MCP server provides 6 tools with basic but incomplete definitions. All tools have descriptions and input schemas are present with proper JSON Schema structure. However, multiple critical gaps reduce quality: (1) Parameter descriptions are missing entirely for most parameters, only 'query' in execute_sql and 'table' in describe_table have descriptions; (2) Output schemas are completely undocumented, callers don't know what fields to expect from results; (3) Error handling is minimal with no guidance on recovery steps; (4) No parameter validation constraints (enums, ranges, patterns); (5) No input parameter descriptions for 4 parameters across tools (database, table in list_tables context). The tool naming is verb-forward and clear (execute_sql, list_tables, describe_table, etc.), which is good. The descriptions are adequate but generic (e.g., 'Execute a SQL query on the MariaDB database' doesn't explain when to use it vs alternatives, prerequisites, or side effects). This server lands in the C/D range, functional for basic use but lacks the robustness expected of production-grade tools.
Show the structure of a table
Execute a SQL query on the MariaDB database
Get the name of the currently selected database
List all available databases
List all tables in the current database
Switch to a different database. Must be called before executing queries.
Output schemas completely undocumented for all tools. LLMs cannot plan downstream operations or extract required fields without knowing the response structure.
Missing parameter descriptions for 'database' (switch_database), and lack of output documentation makes it unclear what format the response takes or whether callers get structured metadata back.
execute_sql description does not state that this tool modifies state (INSERT/UPDATE/DELETE) or warn about side effects and idempotence. LLMs need this distinction to determine retry safety.
No input validation constraints on parameters: 'query' (execute_sql) has no length limit or SQL injection safeguards in the schema; 'table' and 'database' have no format constraints or length bounds. LLMs can pass arbitrary values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 42 | - | v1 |
No pagination support documented for list_tables and list_databases. Large result sets could exceed context window. Tools should accept limit and offset/cursor parameters and document result caps.
Error messages exist in the code (e.g., ERROR_MESSAGES.NO_DATABASE) but are not documented in tool descriptions. LLMs don't know what errors to expect or how to recover. Error handling lacks recovery guidance ('try switch_database first' should be in the description).
Tool descriptions lack dependency and sequencing guidance. Users must know to call switch_database before other tools, but this is only partially documented and not consistently called out for all affected tools.