# FitLLM

Checks whether a local LLM fits in the memory of your GPU, multi-GPU rig or Mac, with the VRAM and KV-cache math behind the verdict.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · ★ 8
- Author: click6067-ship-it
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/click6067-ship-it/fitllm-engine
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/mcp/fitllm

## What it does

FitLLM answers one question: will this model fit? You give it a model and your hardware, it works out weight and KV-cache memory from the model's architecture, and it returns a verdict. The hardware can be a single GPU, several GPUs together, or an Apple Silicon Mac.

It connects as a remote MCP server over Streamable HTTP, so nothing is installed locally. Three tools sit behind it: `check_llm_fit` handles one model and suggests a change when it falls short, `what_fits_on_hardware` ranks the models your machine can run, and `list_supported` shows the model and hardware names the checker knows.

The server is read-only, keeps no state and asks for no login. The flip side is that model and hardware names go to `fitllm.run`. If the agent can't match a name, call `list_supported` first.

## Requirements
- Remote server: `https://fitllm.run/api/mcp`

## Install

### Claude Code (native platform)

**Platform CLI: remote server**
```bash
claude mcp add --transport http fitllm https://fitllm.run/api/mcp
```

### Codex

**Platform CLI: remote server**
```bash
codex mcp add fitllm --url https://fitllm.run/api/mcp
```

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "fitllm": {
            "url": "https://fitllm.run/api/mcp"
        }
    }
}
```

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "fitllm": {
            "type": "http",
            "url": "https://fitllm.run/api/mcp"
        }
    }
}
```

### Gemini CLI

**Platform CLI: remote server**
```bash
gemini mcp add -t http fitllm https://fitllm.run/api/mcp
```

## When to use it
- check whether a model fits before you download it
- see which local models your Mac can run
- find out what to change when a model doesn't fit
- estimate memory on a rig with several GPUs
- look up the model and GPU names the checker accepts

## Example prompts
- Ask fitllm whether Gemma 4 12b fits on an RTX 4090
- Use fitllm to pick models that will run on my Mac
- Call `list_supported` from fitllm and show me the GPU names it knows

## Similar
- [MCP Toolbox for Databases](https://affpapa.org/en/nevibecode/mcp/googleapis-mcp-toolbox-for-databases.md) — MCP server · ★ 16730
- [labelhead-artist-momentum](https://affpapa.org/en/nevibecode/mcp/labelhead-artist-momentum.md) — MCP server · ★ 107701
- [Agent Skills Search Server](https://affpapa.org/en/nevibecode/mcp/agent-skills-search-server.md) — MCP server · ★ 26673
- [Skyvern](https://affpapa.org/en/nevibecode/mcp/skyvern.md) — MCP server · ★ 23283
- [Nhost](https://affpapa.org/en/nevibecode/mcp/nhost.md) — MCP server · ★ 9369
- [PraisonAI](https://affpapa.org/en/nevibecode/mcp/praisonai.md) — MCP server · ★ 9298

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md