Loading ToolBench
Minimal MCP server for real-time control of Synthesizer V Studio 2 via a generic SV-API dispatcher.
SV Copilot defines 16 tools with complete JSON schemas and descriptions. Naming follows verb_noun convention (sv_snapshot_*, sv_edit_*, sv_plan_*, sv_analyze_*, sv_commit_*, sv_compare_*), which is clear and action-oriented. However, descriptions are terse (avg ~80 chars, well below the 194-char baseline for A+ tools), lacking context on WHEN to use each tool or dependencies between them. Schemas are well-structured with type constraints and minLength/minimum bounds, but parameter descriptions are sparse, many lack guidance on expected formats or valid ranges. Output schemas are not documented in the tool definitions. Error handling and recovery guidance are absent. The artifact-based context model (contextId references) is sophisticated but underdocumented for LLM reasoning.
Analyze phrase structure with segmentation and harmonic analysis
Analyze pitch technique characteristics and patterns
Analyze style profile for vocal characteristics and expression patterns
Read or manage artifacts (large results, plans, analysis data)
Commit a generated plan to apply changes to the project
Compare computed pitch between two contexts with anomaly detection
Edit note phrase structure with operations like insert, delete, replace
Descriptions are terse (avg ~80 chars vs 194-char baseline). Most lack WHEN/WHY guidance, prerequisites, or dependencies. E.g., 'Capture a snapshot of a note group' does not explain when to call this vs sv_snapshot_range, or that contextId is required for downstream tools.
Output schemas are not documented. LLMs cannot infer what fields are returned or plan downstream tool calls. E.g., sv_snapshot_group returns a contextId, but this is not declared in the tool definition.
Parameter descriptions are minimal or absent. 'include' arrays lack guidance on valid values. 'operations' in sv_edit_phrase is described as 'List of note structure operations' but does not explain the operation schema or valid types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Patch pitch control points and curves for note groups
Generate an expression plan with emotion, genre, and technique intents
Generate a harmony plan with interval, key, and mode specifications
Generate a lyric alignment plan for phoneme and timing adjustment
Generate a pitch correction plan for computed pitch adjustment
Generate a pitch gesture plan with transient intents and tail policies
Generate a quantize plan for note timing adjustment
Capture a snapshot of a note group with optional filtering and analysis
Capture a range snapshot with musical time boundaries and optional filtering
No error handling or recovery guidance. Tools do not document what errors are possible, whether they are retryable, or what the LLM should do next. E.g., if contextId is invalid, what should the agent do?
Artifact-based context model (contextId) is underdocumented. LLMs cannot reason about artifact lifecycle, validity windows, or when to re-snapshot. No guidance on whether contextId persists across sessions or is ephemeral.