AI Agent Skills Overview
Muapi publishes an agent skill: a small instruction set that coding agents load on their own. Once installed, you ask for what you want in plain language and the agent finds the right model, reads its parameters, uploads your files, runs the task and saves the result into your project.
The muapi-models skill
Gives your agent access to 500+ Muapi models for image, video, audio and speech, lipsync, 3D, LLM text and data tools. The agent:
- Searches the live model catalog for matching models.
- Reads each model's parameters, allowed values, defaults and price.
- Uploads local input files and passes the hosted URLs.
- Submits the task and polls until it finishes.
- Downloads the outputs to the location you name.
Model names are read live, so new models work immediately without updating the skill.
Set up
-
Install the skill (requires Node.js):
npx skills add https://muapi.ai -
Create an API key at muapi.ai/dashboard and export it:
export MUAPI_API_KEY=your_key -
Start your agent from that same shell, then ask for something, for example: "Make a 1:1 product photo of a white ceramic mug with Muapi and save it to ./public/mug.png".
Rerun the install command at any time to update the skill.
Key points
- Authentication is the
x-api-keyheader, read fromMUAPI_API_KEY. Keep the key in your environment, never in project files or chat. - Usage is billed in Muapi credits. See Credits.
- Generated files are kept for 30 days, so the skill tells the agent to save results locally.
- Want to test without spending credits? Create a sandbox key (
POST /api/v1/keyswith{"is_test": true}). It returns mock outputs instantly.
Pages in this section
- Install muapi-models
- What your agent can do
- Run coding agents on Muapi: use Claude Code with your Muapi credits, including unrestricted models
- Claude Code
- Codex CLI
- Troubleshooting
- Changelog
Prefer MCP? See MCP Server. Prefer a terminal? See MuAPI CLI. Looking for the earlier GitHub-based script library? See Agent Skills.