MCP (Model Context Protocol) server that exposes SOUNDGREP music production capabilities as tools for Claude Desktop. Implements stdio-based JSON-RPC MCP protocol for sample search, audio analysis, MIDI generation, stem separation, effects processing, and Ableton Live control.
SOUNDGREP has 17 tools with schemas and descriptions present, but quality is inconsistent. Naming is generally verb-first (search_, analyze_, generate_, apply_, detect_, transcribe_, master_), which is good. However, descriptions are often terse (10-50 chars) and lack actionable context for LLM selection. Many parameters lack descriptions entirely or are vague. Output schemas are not documented. Error handling guidance is absent. The server mixes read-only and write operations without clear idempotency or confirmation patterns for destructive tools (separate_stems, apply_effects, master_audio, generate_*). No tool annotations (readOnlyHint, destructiveHint) are visible in the code.
Create a new track in Ableton Live.
Start playback in Ableton Live.
Set the tempo in Ableton Live.
Stop playback in Ableton Live.
Analyze an audio file and return features: BPM, key, loudness, spectral characteristics.
Analyze song structure (verse, chorus, bridge, etc.) from an audio file.
Apply an effects chain to an audio file (reverb, delay, compression, EQ, etc.).
Descriptions are too brief (10-50 chars) and lack context for LLM tool selection. E.g., 'Stop playback in Ableton Live' (35 chars) does not explain when to use this vs other tools or what happens on failure.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields. E.g., search_samples returns results but structure is unknown, does it return [id, name, url, duration]? This forces LLMs to guess.
Destructive/write tools (separate_stems, apply_effects, master_audio, generate_*) lack confirmation or dry-run patterns. Agents can accidentally overwrite audio files or generate unwanted content without safeguards.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Detect chords from an audio file, returning timestamped chord labels.
Search Freesound.org for audio samples online.
Generate a drum pattern as MIDI.
Generate a MIDI melody based on musical parameters.
Generate a full song with vocals via ACE-Step.
Auto-master a track to match a reference track's tonal and loudness profile.
Search the SOUNDGREP sample library by text description (semantic search via CLAP embeddings).
Separate an audio file into individual stems (vocals, drums, bass, other) using Demucs.
Get the current status of the SOUNDGREP system: running tasks, indexed samples, Ableton connection state.
Transcribe speech or vocals from an audio file using Whisper.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible in code. LLMs cannot distinguish safe read-only tools from destructive ones, risking unintended side effects.
Parameter descriptions are missing or vague. E.g., 'effects' in apply_effects is described as 'List of effect names to apply in order' but does not specify valid effect names, order constraints, or what happens if an invalid effect is passed.