Loading ToolBench
Local-first MCP server for meeting recordings: transcripts, summaries, action items, and decisions. Runs on your machine with whisper.cpp and Ollama — no cloud.
Mixed quality across 9 tools. Naming is verb-forward and clear (normalize_audio, transcribe_audio, extract_action_items). Descriptions are present but brief (avg ~80 chars, below 194-char baseline). Input schemas visible for all tools with type definitions and parameter descriptions. (2) Output schemas documented in CLI tools (Zod) but not visible in MCP server registration. (3) No error handling guidance in descriptions, tools do not explain recovery paths. (4) Optional 'model' parameter in analysis tools lacks enum constraint or format guidance. (5) process_recording and search_meetings descriptions lack detail on what 'full pipeline' and 'ranked results' mean.
Extract action items from a meeting transcript. Returns a JSON list of tasks with owner and due date.
Extract confirmed decisions from a meeting transcript. Returns a JSON list of decisions.
Get detailed information about a specific meeting including transcript, summary, action items, and decisions.
List all processed meetings with their metadata.
Normalize an audio file to 16kHz mono PCM WAV format required by Whisper.
Run the full Samuraizer pipeline on an audio file. Returns summary, action items, decisions, and output file paths.
Search meetings by summary text, name, action items, and decisions. Transcripts are not searched — use get_meeting for those. Returns ranked results with snippets.
Tool definitions in MCP server not visible in source. Tools inferred from file paths and descriptions only. Explicit tool registration with schemas in packages/mcp-server/src/index.ts not provided, preventing verification of actual MCP schema compliance.
Descriptions are too brief (avg 50-60 chars vs 194-char baseline). 'Generate a concise meeting summary' and 'Extract confirmed decisions' lack context on when to use each tool vs alternatives, and do not explain output structure or prerequisites.
Optional 'model' parameter in summarize_transcript, extract_action_items, extract_decisions lacks enum constraint or format guidance. Description says 'Ollama model to use (optional)' but does not list valid models or explain default behavior if omitted.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2025-06-18+ | v2 |
Generate a concise meeting summary from transcript text.
Transcribe an audio file using whisper.cpp. Returns the transcript text.
No error handling guidance in any tool description. Tools do not explain what happens on failure (e.g., if audio file is corrupted, Ollama is unreachable, or meeting not found). LLM has no recovery path.
Output schemas not documented in tool descriptions. CLI tools define Zod schemas (e.g., actionItemSchema with text, owner, dueDate fields) but MCP descriptions do not state what fields are returned, forcing LLM to guess structure.