Bidirectional, idempotent sync of contacts/deals between a local SQLite source-of-truth and CRM APIs (HubSpot/Pipedrive/Salesforce) via one config.
Single tool 'crmsync_scan' has a reasonable description (127 chars, within 10-1024 range) but critically lacks a proper input schema. The FastMCP decorator shows only 'target: str' with no JSON Schema definition visible in the code. Parameter 'target' has no description explaining what it accepts (path? identifier? format?). No output schema documented. Tool name lacks action verb prefix (should be 'scan_crm' or 'perform_crm_scan'). No error handling guidance, no pagination support, no field-level documentation. The description mentions 'Returns JSON findings' but the actual response structure is not documented. This is a minimal implementation that would confuse LLMs about parameter semantics and expected output shape.
Bidirectional, idempotent sync of contacts/deals between a local SQLite source-of-truth and CRM APIs (HubSpot/Pipedrive/Salesforce) via one config.. Returns JSON findings.
No input schema visible. Parameter 'target' has no type definition, description, or constraints in the source code. FastMCP decorator shows bare 'target: str' with no JSON Schema structure.
Tool name 'crmsync_scan' does not start with a clear action verb. Should be 'scan_crm_sync', 'perform_sync', or 'sync_crm' to signal intent. Current name is ambiguous about what action occurs.
Parameter 'target' lacks description. LLMs cannot infer whether it expects a file path, directory, config identifier, or CRM account name. No format, constraints, or examples provided.
Output schema not documented. Description says 'Returns JSON findings' but the actual structure (fields, types, nesting) is not specified. LLMs cannot plan downstream operations without knowing response shape.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
No error handling guidance. Tool description does not explain what happens on invalid target, missing config, API failures, or permission errors. No recovery hints for LLMs.