A skill/client for LLM fallback routing that selects between primary and backup models based on failure status
This server has a single tool 'get_route' with a minimal implementation. While the tool name follows verb_noun convention (get_route), the definition is severely under-specified. The description is 97 characters but lacks critical context about WHEN to use this tool, WHAT it actually does beyond the literal statement, and HOW an LLM should reason about backup model selection. The parameter 'primary_failed' has a type (boolean) and basic description, but there is no documented output schema visible in the source. The tool appears to be a simple utility for model routing but provides no guidance on error handling, retry logic, or how backup routing failures should be handled. There is no evidence of MCP server registration code, tool schema definition, or protocol handlers, only a client class with a method. This suggests the source code provided is incomplete or not the actual MCP server implementation.
Routes to backup model if primary failed, otherwise routes to primary model
No visible MCP server implementation or tool registration. Source shows only a Python class, not an MCP server.
Tool description (97 chars) lacks context about WHEN to call this tool, WHAT constitutes a 'primary failure', and HOW the LLM should decide between models. No guidance on fallback behavior or error recovery.
No documented output schema visible. Return type is inferred as dict (from example_usage.py showing {"backup_model": ...}) but not formally defined with JSON Schema for MCP consumption.
No error handling or recovery guidance. What happens if both primary and backup models fail? Should the LLM retry, call a different tool, or ask the user?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
Tool logic is trivial (single boolean → static string selection). No validation, no edge cases, no idempotency guarantees, no audit trail. Questionable whether this warrants a separate tool vs. being embedded in agent logic.
Parameter 'primary_failed' description is generic ('Whether the primary model failed') and doesn't explain what constitutes failure, how it's detected, or what the LLM should pass.