UVR maestro
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| UVR_MAESTRO_UVR_DIR | No | Explicit desktop/source installation root containing models/. | |
| UVR_MAESTRO_DATA_DIR | No | Override Maestro's owned data root. Default is %LOCALAPPDATA%/uvr-maestro on Windows, ~/Library/Application Support/uvr-maestro on macOS and $XDG_DATA_HOME/uvr-maestro or ~/.local/share/uvr-maestro on Linux. | |
| UVR_MAESTRO_MODELS_DIR | No | Override the managed model root. | |
| UVR_MAESTRO_RUNTIME_DIR | No | Override the managed engine runtime root. | |
| UVR_MAESTRO_ENGINE_PYTHON | No | Development/expert override for an already provisioned interpreter. Setup preflight still verifies it. | |
| UVR_MAESTRO_SETTINGS_ROOT | No | Override the independent profile directory. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| uvr_statusA | Report engine readiness, model roots and last measured backend support. |
| list_modelsA | Read the offline catalog and current disk state without downloading. |
| find_modelC | Rank compatible supported models and expose the task-based selection. |
| refresh_catalogA | Fetch upstream metadata and atomically replace the validated offline snapshot. |
| download_modelB | Download a complete model artifact set into Maestro-owned storage. |
| register_modelA | Register or replace a local model without moving its artifacts or refreshing upstream. |
| unregister_modelA | Remove a local registration while preserving all original weights and configuration. |
| download_missing_modelsC | Plan or download missing models for one supported architecture. |
| remove_modelA | Remove owned artifacts while preserving reused files and shared checkpoints. |
| verify_catalogB | Refresh installation state, check trusted hashes and report orphan weights. |
| get_settingsC | Read defaults overlaid by the selected persisted profile. |
| set_settingsA | Validate and merge profile changes, refusing to write without confirmation. |
| reset_settingsA | Plan or delete a Maestro profile so future runs use bundled defaults. |
| start_separationA | Start one separation and return a queued job. Poll get_job for verified files. |
| separate_audioA | Run one separation and await verified outputs. Use start_separation for long work. |
| get_jobA | Read a job's state, progress, actual outputs and durable manifest path. |
| list_jobsA | List the latest hundred persisted and active jobs. |
| change_journalA | Read the latest fifty confirmed catalog/profile change records. |
| cancel_jobA | Stop queued work or terminate its native worker, preserving partial-run evidence. |
| separate_ensembleB | Queue actual model runs and combine matching sources. Poll get_job. |
| separate_chainB | Queue a cleanup/separation chain using each declared source as the next input. |
| separate_vocal_splitterB | Queue UVR's lead/backing vocal splitter using a vocal model and a BVE model. |
| separate_batchC | Plan or queue per-file separation with aggregate results and cancellation. |
| probe_audioA | Read file metadata without decoding the whole recording. |
| verify_audioB | Decode the complete file in blocks and reject empty or non-finite samples. |
| create_sampleC | Queue a real preview clip in a unique output directory. |
| pitch_shiftB | Queue Rubber Band pitch shifting while preserving duration. |
| time_stretchA | Queue Rubber Band time stretching while preserving pitch. |
| align_inputsB | Queue sample-offset estimation and write the target aligned to the reference. |
| matchering_inputsB | Queue reference mastering with Matchering and write a verified 24-bit WAV. |
| analyze_audioB | Queue full-file peak, RMS, clipping and BS.1770 integrated loudness analysis. |
| check_updateA | Compare the pinned UVR patch with current platform-specific upstream metadata. |
| list_updatesB | Read full-installer URLs and patch source without downloading an installer. |
| uvr_skillsA | List client-neutral workflow guidance available through uvr://skill/{name}. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| separate_stems | Guide a local separation from readiness through verified outputs. |
| restore_recording | Guide a source-routed restoration chain with actual output labels. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| catalog_resource | Read the compact offline model map and lookup routes. |
| settings_resource | Read the current profile and its typed schema. |
| status_resource | Read readiness without requiring a loaded engine. |
TDQS
Scored across 34 tools
Most tools target distinct operations, but the separation family (separate_ensemble, start_separation, separate_audio, separate_chain, separate_vocal_splitter, separate_batch) has overlapping boundaries that require close reading to choose correctly. Model-management tools also overlap somewhat (download_model, download_missing_models, register_model, refresh_catalog), though descriptions provide enough context to disambiguate.
The tool set is predominantly snake_case verb_noun (list_models, download_model, cancel_job, probe_audio), which is readable and predictable. Minor deviations include noun-style names such as uvr_status, change_journal, and matchering_inputs, plus uvr_skills.
34 tools is heavy for an audio separation server, spanning model catalog management, job control, core separation, and several niche audio utilities. Many specialized tools could be consolidated or moved into documented workflows, making the surface feel overextended rather than tightly scoped.
The surface covers model lifecycle, job execution and cancellation, settings, catalog verification, separation workflows, and audio analysis/processing utilities. Minor gaps may exist around detailed per-model inspection or richer job-output retrieval, but most agent workflows are supported end to end.