Koskamo

Traintools MCP

A local, read-only MCP server providing agent-ready PyTorch training diagnostics. It exposes tools for gradient health monitoring, gradient noise scale analysis, batch inspection, data quality checks, plasticity probing, and early stopping recommendations to help debug and optimize deep learning training runs.

Package and access

Local package with source-attributed installation metadata. Koskamo does not execute packages during review.

Traintools MCP logo, by AparajeetSCommunity listingLocal stdioSource reviewed; not executedPublished by AparajeetS
You need:pip install " traintools[mcp] "
Vendor documentation

What you can do with it

  • Diagnose gradient health issues like vanishing or exploding gradients
  • Check if batch size is wasting compute using Gradient Noise Scale
  • Inspect batches for imbalance, errors, or scale issues
  • Determine if training should stop early based on plasticity probes

Source and maintenance

Community listingRegistry 0.6.2
mcp-registry: io.github.AparajeetS/traintools@0.6.2Source repositoryProject website
Maintenance checked

Packages and installation

Source-provided package data. Koskamo has not executed or live-probed this package.

traintools 0.6.2

pypiuvx
traintools package definition
{
  "version": "0.6.2",
  "transport": "stdio",
  "identifier": "traintools",
  "runtimeHint": "uvx",
  "registryType": "pypi",
  "packageArguments": [
    {
      "type": "positional",
      "value": "mcp"
    }
  ],
  "runtimeArguments": [
    {
      "name": "--with",
      "type": "named",
      "value": "mcp"
    }
  ]
}

Tools

Reported by the attributed source; not independently observed by Koskamo.

No tool definitions available.

Connect from your client

Each snippet configures the same source-provided package; replace placeholders before use.

This server is operated by the named third party. Listing and endpoint checks are not an endorsement or a security audit. Review the operator's permissions, terms, and data handling before connecting. How listings are checked and corrected.

Cursor

Merge this local stdio configuration, replace every placeholder environment value, and restart Cursor. Koskamo has not executed this package.
Cursor configuration
{
  "mcpServers": {
    "traintools-mcp": {
      "command": "uvx",
      "args": [
        "traintools==0.6.2",
        "mcp"
      ]
    }
  }
}

Cline

Merge this local stdio configuration, replace every placeholder environment value, and restart Cline. Koskamo has not executed this package.
Cline configuration
{
  "mcpServers": {
    "traintools-mcp": {
      "command": "uvx",
      "args": [
        "traintools==0.6.2",
        "mcp"
      ]
    }
  }
}

VS Code

Merge this local stdio configuration, replace every placeholder environment value, and restart VS Code. Koskamo has not executed this package.
VS Code configuration
{
  "servers": {
    "traintools-mcp": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "traintools==0.6.2",
        "mcp"
      ]
    }
  }
}

Claude Desktop

Merge this local stdio configuration, replace every placeholder environment value, and restart Claude Desktop. Koskamo has not executed this package.
Claude Desktop configuration
{
  "mcpServers": {
    "traintools-mcp": {
      "command": "uvx",
      "args": [
        "traintools==0.6.2",
        "mcp"
      ]
    }
  }
}

Let your agent find its tools

Find servers like Traintools MCP from your agent. Add Koskamo once to search the index for endpoints, access requirements, and client setup. Public discovery needs no key.

Connect Koskamo Explore workflows
See all 3,755