mcp400 exposes three database schema inspection tools for AS/400 systems. All tools have names following verb_noun convention (get_*), descriptions are present but minimal (34-71 chars, below the 50-200 char LLM-optimized range), input parameters have types and descriptions, but output schemas are NOT documented, a critical gap. Tools are read-only and safely composed but lack error handling guidance, validation examples, and any depth of context around when/why to call each tool vs others. The server uses Spring AI framework with @Tool annotations, indicating proper MCP registration, but transport is UNKNOWN (likely STDIO based on Spring Boot starter pattern), which caps protocolReadiness severely.
Gets a list of all tables(name and description) in a specific schema on the database/as400/ibmi
Gets a list of all columns(name, description, datatype), index and primary key by schema and tableName
Gets a list of schemas containing the table
Output schemas are not documented. MCP400Service returns Table, TableInformation, and List objects, but the MCP tool definitions do not include output schema declarations. LLMs cannot infer what fields to expect from these tools.
Tool descriptions are too brief and lack LLM-optimization. Descriptions range 34-71 chars, falling below the 50-200 char optimal range for LLM selection. Missing WHAT/WHEN/WHY context. E.g., 'Gets a list of all tables...' does not explain when to call this vs mcp400_get_schemas_for_table or what the LLM should do with the results.
No error handling or recovery guidance. If a schema does not exist or the database is unreachable, the tools will raise SQLException. No tool descriptions guide the LLM on what to do: retry? Ask the user? Call a different tool? SQLException stack traces are not actionable for agents.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 49 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
Input parameters lack format/constraint documentation. 'schema' and 'tableName' are described as strings but do not specify: are they case-sensitive? Do they accept wildcards? The code uppercases schema/tableName internally, but this is not documented in the parameter descriptions. LLMs may pass lowercase or invalid names.
No pagination support declared. If a schema contains hundreds of tables, mcp400_get_all_tables_for_schema will return them all, potentially exhausting the LLM context window. The tool should document a result limit and offer pagination (limit, offset/cursor).
Composition ambiguity: three tools all query different aspects of the same schema/table structure. Tool descriptions do not clarify when to call get_schemas_for_table (find which schemas contain a table) vs get_all_tables_for_schema (list all tables in a schema). LLMs may pick the wrong tool.