An Elixir/Phoenix-based MCP server that provides pooling and management of AI model serving infrastructure with tool dispatch, registry-based MCP tool catalog, and gateway payloads handling
This MCP server has CRITICAL failures across all definition quality dimensions. The tool registry in lib/codex_pooler/mcp/tool_registry.ex defines 5 tools, but they are tool FAMILIES with metadata-only descriptions, not actual tool implementations. The descriptions state 'metadata only, actual tools defined in family module', meaning the actual schema, parameter definitions, and tool descriptions are not visible in the provided source code. Per the hard scoring rules, if tool definitions cannot be verified in source, they are capped at 50. No input schemas are visible for any tool. No parameter descriptions are visible. No output schemas are documented. Error handling patterns are not evident. The server has toolAnnotations enabled but no evidence of how they are applied to actual tools. This is a bare-bones, incomplete MCP server definition.
MCP tool family from CodexPooler.MCP.Tools.Foundation - metadata only, actual tools defined in family module
MCP tool family from CodexPooler.MCP.Tools.LogMetadata - metadata only, actual tools defined in family module
MCP tool family from CodexPooler.MCP.Tools.OperatorMetadata - metadata only, actual tools defined in family module
MCP tool family from CodexPooler.MCP.Tools.PoolMetadata - metadata only, actual tools defined in family module
MCP tool family from CodexPooler.MCP.Tools.QuotaMetadata - metadata only, actual tools defined in family module
No tool schemas visible in source code. All 5 tools are registered as family placeholders with no input/output schema definitions shown.
Tool descriptions are metadata-only placeholders ('MCP tool family from ... - metadata only, actual tools defined in family module'). These do not explain what the tool does, when to use it, or what it returns.
Tool names use '_tool_family' suffix and are generic (foundation, pool_metadata, quota_metadata, operator_metadata, log_metadata). No action verbs (get_, create_, list_, search_). LLMs cannot infer intent from these names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 11 | 2025-06-18+ | v2 |
No parameter descriptions visible in source. Without parameter-level documentation, LLMs cannot understand what each input means or how to validate it.
No error handling guidance visible. No recovery messages, error classification, or self-correction paths for LLMs.
Tool definitions are inferred (tool family registration visible, but actual tool schemas and parameters are deferred to submodules not shown in source).