TruePath Recorder MCP has 4 tools with basic naming and descriptions, but critical gaps in schema completeness and parameter documentation. Tool names follow verb_noun convention (get_status, list_sources, start_recording, stop_recording), which is good. However, input schemas for get_status, list_sources, and stop_recording are empty objects with no properties defined. The start_recording tool has a schema with 6 parameters, but 4 of them lack descriptions in the schema itself (only present in the rubric summary, not in the actual code). Output schemas are completely undocumented, LLMs cannot predict what fields to expect from responses. Error handling is minimal: generic 'No response' and 'Unknown tool' messages provide no recovery guidance. The code shows synchronous HTTP calls to a localhost control API, but the tool definitions do not document this dependency or what happens if the app is unreachable.

• Add complete input schemas to get_status, list_sources, and stop_recording. Even if they accept no parameters, explicitly define 'properties: {}' and 'required: []' in the schema.
• Add descriptions to all 6 parameters in start_recording's schema. Example: 'system_audio': {'type': 'boolean', 'description': 'Include system audio in the recording.'}
• Document output schemas in tool descriptions. Example for list_sources: 'Returns an object with displays (array of {id: int, title: string}) and windows (array of {id: int, app: string, title: string}).'
• Replace generic error messages with actionable LLM guidance. Example: 'TruePath Recorder unreachable (retryable). Ensure the app is running and Settings → AI Agent Control is enabled. Retry in 2 seconds.' Include error classification (retryable/user-fixable/fatal).
• Document parameter dependencies in start_recording descriptions. Example for display_id: 'Display ID to capture (from list_sources). Only valid when mode=display. Ignored for mode=window or mode=area.'