# Unsloth Mcp Server

> MCP server for LLM fine-tuning with Unsloth. 33 tools, 180 tests, RunPod GPU integration. Fine-tune 2x faster with 80% less memory.

- **Type:** MCP server
- **Install:** `agentstack add mcp-scientiacapital-unsloth-mcp-server`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [ScientiaCapital](https://agentstack.voostack.com/s/scientiacapital)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** Apache-2.0
- **Upstream author:** [ScientiaCapital](https://github.com/ScientiaCapital)
- **Source:** https://github.com/ScientiaCapital/unsloth-mcp-server

## Install

```sh
agentstack add mcp-scientiacapital-unsloth-mcp-server
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Unsloth MCP Server

> **33 MCP tools** for LLM fine-tuning via Claude. 180 tests passing.
> Integrates Unsloth (2x faster, 80% less memory) with RunPod GPU management.

---

## What It Does

- **33 MCP tools** for model fine-tuning, tokenizer training, GPU management
- **RunPod integration** - create, start, stop pods programmatically
- **Knowledge capture pipeline** - OCR, AI enhancement, training data generation
- **Checkpoint management** - save/load/resume training sessions
- **Cost tracking** - budgets, alerts, expense reporting

## Goals

Enable Claude to orchestrate LLM fine-tuning workflows without manual Python scripting.

## Quick Start

```bash
cd unsloth-mcp-server
npm install && npm run build
npm test  # 180 tests passing
```

Add to Claude Code MCP settings:

```json
{
  "mcpServers": {
    "unsloth-server": {
      "command": "node",
      "args": ["/path/to/build/index.js"]
    }
  }
}
```

## Current Status

| Component             | Status                     |
| --------------------- | -------------------------- |
| MCP tools (33)        | Working                    |
| Test suite (180)      | Passing                    |
| RunPod pod management | Working                    |
| Knowledge capture     | Working                    |
| Checkpoint management | Working                    |
| GRPO training         | Dtype compatibility issues |
| SFT training          | Environment-dependent      |

## Known Issues

- GRPO: Dtype mismatches in some configurations
- Requires Python 3.10-3.12 (not 3.13)
- GPU required for fine-tuning

## GTME Skills Developed

Building toward Go-To-Market Engineer through hands-on projects:

| Skill Area                    | What I Learned                                                                 |
| ----------------------------- | ------------------------------------------------------------------------------ |
| **Developer tooling**         | Built MCP server that developers actually use - learned API design for DX      |
| **Cost optimization**         | Implemented budget tracking, alerts - understand unit economics of GPU compute |
| **Infrastructure automation** | RunPod API integration - programmatic cloud GPU provisioning                   |
| **Testing discipline**        | 180 tests with Jest - shipping quality that earns trust                        |
| **Technical documentation**   | Writing docs that reduce support burden                                        |
| **Product iteration**         | v2.0 → v2.3.0 based on real usage feedback                                     |

## Tech Stack

Node.js/TypeScript, MCP SDK, Python/Unsloth, Jest, RunPod API

## Requirements

- Node.js 18+
- Python 3.10-3.12
- NVIDIA GPU with CUDA (for training)

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [ScientiaCapital](https://github.com/ScientiaCapital)
- **Source:** [ScientiaCapital/unsloth-mcp-server](https://github.com/ScientiaCapital/unsloth-mcp-server)
- **License:** Apache-2.0

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-scientiacapital-unsloth-mcp-server
- Seller: https://agentstack.voostack.com/s/scientiacapital
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.