Agent work can now live in a persistent page with its own backend, callable by the agent through hosted MCP, instead of scattered chat threads. If you build agent plugins, this is a new packaging and runtime model.
Key takeaways · AI-distilled
MiniMax Code's Mini Apps are plugins with a miniapp.json manifest, a frontend and a Node.js server that exposes features to the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → through hosted MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition →, so an agent's work stays on a live page instead of being scattered across conversations.
The host owns processes, ports, authentication, credentials and rollback-capable publishing. Plugins and pages cannot read account credentials, and connectors and agent permissions must be declared in the manifest before use.
The /miniapp-creator command first runs a read-only query against each data source to confirm it returns real data, then writes the page and backend, publishes the app, and checks the result with screenshots in the built-in browser.
Anything a page sends to the agent lands in the user's input box for review first. Mini App code still runs locally as the user, so MiniMax says to install packages only from trusted sources. Mini Apps work in local sessions of the macOS and Windows desktop app.
Terms in this piece · Glossary
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
