Local voice/text AI assistant with Claude Agent SDK, Hyprland integration, and Piper TTS. Includes FastAPI backend, WebSocket bridge, wake-word detection, and WhatsApp MCP server.
12 tools with mixed quality. All have descriptions (10-200 chars, mostly adequate) and input schemas with typed parameters. However, critical gaps: (1) No output schemas documented anywhere, LLMs cannot plan downstream calls or extract results. (2) No error handling guidance, tools return raw API responses without recovery hints. (3) Parameter descriptions lack constraints (enums, ranges, formats), e.g., 'limit' has no min/max, 'sort_by' has no enum. (4) No tool annotations (readOnlyHint, destructiveHint) despite clear read/write distinction. (5) Recipient parameter in send_* tools accepts both phone and JID but lacks format guidance. (6) No pagination metadata (total_count, next_cursor) in list_* responses. Naming is solid (verb_noun pattern), but schema completeness and output documentation are severely lacking.
Download media from a WhatsApp message and get the local file path.
Get WhatsApp chat metadata by JID.
Get all WhatsApp chats involving the contact.
Get WhatsApp chat metadata by sender phone number.
Get most recent WhatsApp message involving the contact.
Get context around a specific WhatsApp message.
No output schemas documented. LLMs cannot infer what fields list_messages, get_chat, or send_message return, forcing them to guess at response structure and breaking downstream tool chaining.
No error handling guidance. Tools return raw API errors without recovery hints (e.g., 'User not found. Try search_contacts() first.'). Agents cannot self-correct on failures.
Parameter constraints missing. 'limit' in list_* tools has no min/max bounds (unbounded integers risk timeouts). 'sort_by' in list_chats lacks enum constraint, LLM may hallucinate invalid values like 'date_created'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Get WhatsApp chats matching specified criteria.
Get WhatsApp messages matching specified criteria with optional context.
Search WhatsApp contacts by name or phone number.
Send any audio file as a WhatsApp audio message to the specified recipient. For group messages use the JID. If it errors due to ffmpeg not being installed, use send_file instead.
Send a file such as a picture, raw audio, video or document via WhatsApp to the specified recipient. For group messages use the JID.
Send a WhatsApp message to a person or group. For group chats use the JID.
Recipient parameter in send_* tools accepts 'phone number with country code but no + or other symbols' OR 'JID' but provides no format validation or examples. LLM cannot distinguish valid formats without trial-and-error.
No tool annotations (readOnlyHint, destructiveHint). 8 read-only tools and 4 write tools are unmarked, preventing agents from reasoning about side effects and retry safety.