SaC — Software as Content

What SaC — Software as Content does

SaC — Software as Content MCP server lets an agent respond with a live application that users can open in a browser and interact with. Generated interfaces are intended to change as the conversation continues, so the result is not limited to a static page, screenshot, or block of markdown. The project is currently at version 0.1.2 and is described as alpha software.

The SDK is aimed at tasks where users need to inspect options, manipulate information, or guide the result. Documented examples include trip planning, data-analysis dashboards, comparison shopping, project planning, research, financial reviews, decision aids, and internal tools. Simple questions, one-time automations, and text-only conversations are listed as less suitable cases.

How it works

A conversation produces an app and exposes its URL. The same URL remains associated with the conversation while the agent evolves the existing app instead of creating a separate page for every turn. A user can interact through controls in the app or send a message in chat; both paths return input to the agent through the same callback loop.

The project supports two operating patterns. sac serve starts a local experience with a built-in agent loop, after which the user opens http://localhost:18420. Alternatively, an existing agent can connect through the MCP integration. The Python API provides a SaC object, a conversation, and a generated app whose URL and TSX source can be accessed by code.

Setup and configuration

Install the Python package with pip install sac-sdk. Running sac serve requests an API key on first use and saves it. The README does not specify the key name or the service behind that key.

For Claude Code, run sac setup claude-code to register SaC as an MCP server, then restart Claude Code. Codex and OpenClaw use setup commands that install a skill rather than an MCP registration: sac setup codex or sac setup openclaw. Those integrations also require sac serve to remain running.

Python applications can instantiate SaC and call conversation() followed by generate(). The language-model provider, search provider, and storage implementation are replaceable through the SDK interfaces. The README shows FileStore as one storage option and indicates that prompts and the default design system are stored in the source tree.

Tools and capabilities

• Generate a live app from a natural-language request.
• Evolve an existing app as the conversation continues.
• Accept user input through app controls and chat messages.
• Stream a React-based interface to a local URL.
• Expose the workflow through an MCP stdio server for Claude Code.
• Return an app URL and generated TSX through the Python API.
• Replace model, search, and storage providers with custom implementations.

SaC — Software as Content is therefore most relevant when the interface itself is part of the agent's answer. Its documented architecture includes a runtime generation-and-evolution pipeline, an HTTP layer with FastAPI and SSE streaming, an MCP server, and an iframe-based renderer.

Limitations and notes

The project does not claim to replace ordinary text responses or one-shot automation workflows. Its core protocol is described as stable in production use, but the SDK surface is still being refined toward version 1.0. The README does not document supported API-key providers, environment-variable names, hosted deployment instructions, or a compatibility list for Claude Desktop, Cursor, Windsurf, or Cline.

The repository is licensed under Apache-2.0. Local development uses pip install -e ..

Getting started with this SaC — Software as Content MCP server

Always refer to the official documentation for the most accurate and up-to-date information.