GPU-Gate-MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| GPU_GATE_HOME | No | Status directory (state home). Defaults to %LOCALAPPDATA%\gpu-gate on Windows, otherwise ~/.local/share/gpu-gate. | |
| GPU_GATE_CONFIG | No | Path to the config file. Defaults to <status directory>/config.json. Can also be set with the --config option. |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| comfy_batchA | Run ComfyUI jobs on the GPU while the local LLM is unloaded; blocks until all are done and returns per-job status and output file paths. jobs: {"kind": "workflow", "path": "", "name": ""} or {"kind": "command", "command": ["python", "gen.py", ...], "cwd": "", "timeout_s": }. options (all optional): stop_on_error, job_timeout_s, batch_timeout_s, keep_executor (leave an already running ComfyUI up), force (go on although another LLM server answers), comfy_variant (another start .bat in the ComfyUI folder). |
| voice_batchA | Run VoiceStudio speech jobs on the GPU while the local LLM is unloaded; blocks until all are done and returns per-job status and output file paths. jobs: {"text": "...", "out": "<output .wav path>", "profile_id": ..., "instruct": ..., "speed": ..., "language": ...} (the batch script's job format). options (all optional): batch_timeout_s, keep_executor, force. |
| gate_statusA | Read-only snapshot: the gate's state, the running batch's progress, whether the LLM is loaded, whether ComfyUI/VoiceStudio are up, GPU memory per process, warnings and the last batch result. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 3 tools
The two batch tools are separated by backend (ComfyUI vs VoiceStudio) and gate_status is a clear read-only counterpart, so boundaries are mostly clear. The shared 'batch while LLM unloaded' framing and the 'gate' theme create mild risk of confusion, but the names and descriptions make the target system explicit.
comfy_batch and voice_batch follow a consistent noun_action pattern with a shared _batch suffix, and gate_status is readable in the same style. The minor deviation (status vs batch) is natural and does not hurt predictability.
Three tools is well-scoped for a GPU gate whose purpose is to run two classes of GPU jobs and report state. It is slightly thin—there is no per-job submission or cancellation tool—but nothing is redundant.
Coverage includes batch execution for both backends plus a status snapshot, which fits the stated purpose. However, there is no way to cancel or abort a long-blocking batch, nor to list or inspect past batches beyond the last result, leaving notable gaps for a blocking-execution server.