Guaardvark plugin
A plugin and MCP server that let Claude Code drive the local Guaardvark AI studio on your own graphics card
- 292 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Screen agents click and type in a real browser on a separate Ubuntu desktop
- Once desktop, shell and browser tools are allowlisted, any agent connected over MCP gets them
- The studio stores and processes voice, video and files on your machine
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/guaardvarkDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Screen agents click and type in a real browser on a separate Ubuntu desktop
- Once desktop, shell and browser tools are allowlisted, any agent connected over MCP gets them
- The studio stores and processes voice, video and files on your machine
This is third-party code. Review the repository files before installing.
What it does
The Guaardvark plugin for Claude Code connects the agent to the local Guaardvark AI studio. It asks for the path to your installed studio, starts the MCP server from there and loads 15 skills as /guaardvark:<skill>. The skills cover images, video, music, voice, LoRA training and a swarm of coding agents. For Cursor, Codex, Gemini CLI and other clients, one command writes the same server into their configs. By default the server exposes more than 50 tools: RAG over files, generation, memory, code search and web. Desktop, shell and browser tools stay hidden until you add them to the allowlist. The studio itself runs models and generators on your hardware through Ollama and ComfyUI. It has chat over your own files with source passages shown, screen agents with their own Ubuntu desktop and a swarm of coding agents, each in its own git worktree. Outbound actions such as web access and publishing are off by default and listed in egress.json.
Who it is for. For owners of an NVIDIA graphics card who keep media generation and agents on their own hardware and drive them from a coding agent.
Good fit when
- Claude Code should make a video, a song or a set of images on your graphics card with no cloud keys
- A refactor needs to be split across several parallel coding agents
- You want a local RAG chat over your documents that the agent reaches through MCP
- You need to train a character LoRA from a few photos and keep its look stable across images and video
Not a fit when
- You have no NVIDIA graphics card with enough memory: the video pipeline will not run on other hardware
- You want the plugin without installing the studio: the plugin starts the MCP server from a local Guaardvark copy
- The agent must publish and reach the internet right away: outbound paths are off and you enable them by hand in egress.json
Example request
Run a coding-agent swarm to split this refactor across modules, and merge the results back in dependency orderLimitations
The full video pipeline needs Linux and an NVIDIA card with 16 GB of memory, while the Wan 2.2 5B model fits in 12 GB. Chat, RAG and agents run on smaller hardware, with tiers set out in a separate document. The plugin and the guaardvark package from PyPI do not work without a local studio copy: they need its path. Fixes Guaardvark writes for its own code wait for your review and are never applied on their own. Publishing and voice cloning happen only after explicit consent.
How to disable. Stop the stack with `./stop.sh` in the install folder and delete the ~/guaardvark folder. Remove the plugin with `/plugin uninstall guaardvark@guaardvark` and `/plugin marketplace remove guaardvark`.
FAQ
Do I need a cloud API key for Guaardvark?
No. Models, search and generators run on your hardware through Ollama, ComfyUI and the studio's GPU services, with no per-token fees.
How do I connect Guaardvark to Claude Code?
Run `/plugin marketplace add guaardvark/guaardvark` and `/plugin install guaardvark@guaardvark`. The plugin asks for the path to your installed studio and starts the MCP server from there.
How do I connect Guaardvark to Cursor or Codex?
`python -m backend.mcp install` in the studio folder finds the clients on your machine and writes the guaardvark server into them, backing up the configs. `python -m backend.mcp doctor` diagnoses a broken setup.
How does Guaardvark run a swarm of coding agents?
Each swarm agent works in its own git worktree, and results merge back in dependency order. The swarm runs on Claude Code or fully locally through Ollama.
Can the MCP server run over HTTP?
Yes, `python -m backend.mcp http` starts streamable HTTP at 127.0.0.1:8788/mcp, loopback-only by default. `python -m backend.mcp list-tools` prints the current tool list.
Discussion
No questions or reviews yet.
Related
PluginsMulti-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
A Claude Code plugin marketplace with ready sets of agents, slash commands, hooks and skills, plus a search site
An operating layer for Claude Code and Codex: one /do entry point, project memory across sessions, safety hooks, cost telemetry and parallel agents
AI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
A Claude Code plugin that connects the agent to shared organizational memory with access control, audit and signed erasure
A desktop app where Claude Code, Codex, Gemini and other agents work as a team of specialists in a shared project
README in short
The README opens with four reasons to pick the studio and a quick start for Linux with NVIDIA. A comparison table with similar projects follows, plus links to Docker, macOS and manual installs.