Guaardvark — Claude Code MCP server: local AI studio
Hooks your agent up to a self-hosted studio on your own GPU: chat over your files, plus video, image, music and voice generation. The Guaardvark app must be installed.
claude mcp add guaardvark -- uvx guaardvarkWhat Guaardvark does
This MCP server opens Guaardvark's flows to Claude Code, Codex, Cursor, Copilot and other agents: retrieval over your own documents, agents driving a real desktop, parallel coding agents, and generators for video, images, music and speech. The work happens on your hardware through Ollama, ComfyUI and Guaardvark's GPU services.
The server starts with uvx guaardvark, so uv has to be present. The Guaardvark application itself needs to be running already — the server has nothing to talk to otherwise. Desktop, shell and browser tools stay hidden until you allowlist them, which is a deliberate choice by the authors and caps what an agent can touch on your machine.
The studio targets Linux with an NVIDIA card. Video generation is advised to have 16 GB of VRAM, the Wan 2.2 5B model fits in 12 GB, and chat, RAG and agents get by with less. On a fresh Ubuntu, install curl and git first.
What it helps with
- wire local video and image generation into Claude Code
- search your own documents without shipping them to a cloud
- run parallel coding agents in separate git worktrees
- generate a song or a voiceover straight from the agent chat
- train a character LoRA from a few photos on your own GPU
Example prompts for the agent
- Connect the Guaardvark MCP server and list the tools it exposes
- Make a music video from this track with Guaardvark
- Swarm this refactor with coding agents
How to install Guaardvark
Platform CLI
claude mcp add guaardvark -- uvx guaardvarkuvx runs the command named after the package; if the README names another one, use uvx --from guaardvark <command>
Platform CLI
codex mcp add guaardvark -- uvx guaardvarkuvx runs the command named after the package; if the README names another one, use uvx --from guaardvark <command>
Config file
{
"mcpServers": {
"guaardvark": {
"command": "uvx",
"args": [
"guaardvark"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from guaardvark <command>
Config file
{
"servers": {
"guaardvark": {
"type": "stdio",
"command": "uvx",
"args": [
"guaardvark"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from guaardvark <command>
Platform CLI
gemini mcp add guaardvark uvx guaardvarkuvx runs the command named after the package; if the README names another one, use uvx --from guaardvark <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Guaardvark
How do I install Guaardvark in Claude Code?
Run claude mcp add guaardvark -- uvx guaardvark. You also need the Guaardvark app deployed, since the server connects to it.
What does Guaardvark do?
It is an MCP server for a local AI studio: the agent gets chat over your files, video, image, music and voice generation, and parallel coding agents. Everything runs on your GPU.
What hardware does Guaardvark need?
Linux with an NVIDIA GPU. Video is advised to have 16 GB of VRAM, Wan 2.2 5B needs 12 GB, and chat, RAG and agents run with less.
