Provides tools for training, inspecting, validating, testing, predicting, and checkpointing PyTorch Lightning models.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@PyTorch Lightning MCP Server train a ResNet-18 on CIFAR-10 dataset"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A minimal integration layer exposing PyTorch Lightning via a structured, machine-readable API for tools, agents, and orchestration systems.
Structured APIs for training, inspecting, validating, testing, predicting, and checkpointing models
PyTorch Lightning execution
Stdio and HTTP server modes
Related MCP server: mcp-server-lightning-exec
Python 3.10–3.12
PyTorch Lightning (compatible version)
uv (recommended for dependency management)
curl -Ls https://astral.sh/uv/install.sh | sh
git clone https://github.com/<your-org>/lightning-mcp.git
cd lightning-mcp
uv sync --all-extrasYou can run the MCP server via CLI:
# Stdio server (default)
uv run lightning-mcp
# HTTP server
uv run lightning-mcp --http --host 0.0.0.0 --port 3333echo '{"id":"1","method":"lightning.inspect","params":{"what":"environment"}}' | uv run lightning-mcpcurl -X POST http://localhost:3333/mcp \
-H "Content-Type: application/json" \
-d '{"id":"1","method":"lightning.inspect","params":{"what":"environment"}}'The MCP server exposes the following tools (methods):
lightning.trainTrain a PyTorch Lightning model with explicit configuration.
Input schema:
{
"model": {"_target_": "string", ...},
"trainer": { ... }
}lightning.inspectInspect a model or the runtime environment.
Input schema:
{
"what": "model | environment | summary",
"model": {"_target_": "string", ...} // required for model inspection
}lightning.validateValidate a PyTorch Lightning model.
Input schema:
{
"model": {"_target_": "string", ...},
"trainer": { ... }
}lightning.testTest a PyTorch Lightning model.
Input schema:
{
"model": {"_target_": "string", ...},
"trainer": { ... }
}lightning.predictRun prediction/inference with a PyTorch Lightning model.
Input schema:
{
"model": {"_target_": "string", ...},
"trainer": { ... }
}lightning.checkpointManage model checkpoints: save, load, or list.
Input schema:
{
"action": "save | load | list",
"path": "string", // for save/load
"directory": "string", // for list
"model": { ... } // for save/load
}To list all available tools and their schemas at runtime:
echo '{"id":"1","method":"tools/list","params":{}}' | uv run lightning-mcpuv run pytestSee CONTRIBUTING.md and DEVELOPMENT.md.
Apache 2.0
Manage model checkpoints: save, load, or list.
Inputs: action (required), path, model, directory.
Inspect models or runtime environment.
Inputs: what (required), model.
Run prediction/inference with a PyTorch Lightning model.
Inputs: model (required), trainer.
Test a PyTorch Lightning model.
Inputs: model (required), trainer.
Train a PyTorch Lightning model with explicit configuration.
Inputs: model (required), trainer.
Validate a PyTorch Lightning model.
Inputs: model (required), trainer.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 6 tools
Each tool targets a distinct stage of the PyTorch Lightning model lifecycle (train, validate, test, predict, checkpoint, inspect), with no overlap in functionality.
All tools follow a consistent 'lightning.<action>' pattern (e.g., lightning.train, lightning.test), making it easy to infer purpose from name.
6 tools cover the essential operations for a PyTorch Lightning workflow without being excessive or insufficient.
The set covers the main lifecycle stages, though advanced features like hyperparameter tuning or model export are absent; still satisfactory for core usage.
Discover, compare, route, and execute machine-accessible capabilities for AI agents.
Build, validate, deploy — HTTP APIs, cron jobs, webhooks and MCP tools — from your AI client.
Turn any task into the right API calls: discover, evaluate, and integrate public APIs.
Build, validate, and deploy multi-agent AI solutions from any AI environment.
