Monarch

View saved version →

---
name: monarch-account-observer
description: Read the connected person’s own Monarch account, balance, usage and training status.
---

# Observe Monarch accounts

Use `get_monarch_profile` when account identity matters; use `get_account_balance` and `get_compute_usage` for finances and `list_training_jobs` or `get_training_job` for jobs. Report credits and usage separately. All operations in this skill are read-only.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-compute-advisor
description: Compare current compute capacity and prices without allocating GPUs.
---

# Advise on Monarch GPUs

Use `list_available_gpus` with the relevant GPU family, region or maximum hourly price, then compare returned hardware, capacity and rates. `list_models` and `search_datasets` help assess compatibility. Do not reserve GPUs or start a job in this advisory workflow.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-data-assets
description: Inspect models and training datasets with their available metadata.
---

# Explore model and dataset assets

Use `search_datasets` to find data, `get_dataset` to inspect a specific dataset, and `list_models` or `get_model` for model metadata. Explain readiness, licensing and format only when returned by Monarch. Binary uploads are not supported by these tools.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-external-publishing
description: Help users plan publication while keeping public publishing outside this plugin’s current tool set.
---

# Prepare publication checklists

This public release does not expose any external model or content publishing tool. Help the user identify the artifact, destination, rights and consent required, then explain that publication must be completed in their Monarch account outside this MCP. Never claim to have published anything.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-get-started
description: Connect Monarch and discover the public AI compute workflows the account can actually use.
---

# Get started with Monarch

After OAuth sign-in, use `get_monarch_profile` to verify account identity, `list_available_gpus` for capacity, `list_models` for models, `list_training_jobs` for existing jobs, and `get_account_balance` for credits. Choose an explicit tool for every action. Real model inference uses `run_model_inference`; model training uses `start_training_job` only with informed confirmation.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-projects-runtime
description: Explore existing Monarch Cloud projects and deployment records without altering infrastructure.
---

# Inspect Monarch Cloud projects

Use `list_projects` to select a project, then `get_project` and `list_project_deployments` for details. This plugin exposes project inspection, not project creation, migration, credential rotation or infrastructure deletion.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-serving-inference
description: Inspect available models, test sandbox inference and run approved live inference.
---

# Run Monarch model inference

Use `list_inference_models` or `list_models` to select a model. `run_model_inference` defaults to `sandbox=true`, which provides a simulated response without billable compute. Before setting `sandbox=false`, explain costs and obtain explicit approval, then set `confirm_billable=true`. Inspect live returned status before claiming completion; never automatically retry an uncertain billable request.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-training
description: Start and monitor authorized fine-tuning jobs with explicit cost approval.
---

# Manage Monarch training jobs

Use `list_available_gpus`, `search_datasets` and `list_models` to choose inputs. `list_training_jobs` and `get_training_job` inspect existing training state. Use `start_training_job` only after confirming the model, dataset and expected costs; the tool requires `confirm_billable=true`. `cancel_training_job` requires a specific job ID and explicit `confirm_cancel=true`. If submission is uncertain, inspect jobs before retrying.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: monarch-voice
description: Browse supported speech model families and voice catalog options.
---

# Explore Monarch Voice

Use `list_voice_models` and `list_voices` to inspect available voice models, voices and languages as returned by Monarch. This public MCP does not yet expose speech generation, transcription or voice cloning. Never fabricate an audio output or claim another person consented to voice replication.

## Safety and scope
- All account data and actions require the signed-in Monarch account. Use only the tools explicitly present in this customer MCP; do not invent unlisted operations.
- Never infer compute availability, GPU pricing, deployment status, resource ownership or results that the API has not returned.
- Confirm budget and expected effects before real inference, model training or cancellation. Sandbox inference is simulated rather than a live model completion.
- Do not initiate purchases, top-ups, payouts, external publications or irreversible infrastructure changes from this plugin.
- Do not request or output passwords, credentials, access tokens, or private connection secrets.
- When the requested action is not implemented as a listed tool, state the limitation instead of using a generic executor.

Referenced files: 1 — file references are metadata, not package contents.