Plug your home agent into your kid's iPad
Sprout gives your agent an MCP server and a kid-side runtime. Here is the whole loop, from prompt to receipt.
If you run Claude Code, Codex, or any MCP-capable agent, you already have the brain. What you don't have is a surface your kid opens every day, with the guardrails a kid needs. That is the gap Sprout fills: your agent does the thinking, Sprout runs the kid side.
Here is the loop end to end.
1. Connect
Sprout ships an MCP server. You authenticate once with OAuth from your agent's config, scoped to your family. Your agent can now see what you let it see: your kids' names and ages, their plans, their gem balances.
npx -y mcp-remote https://api.sproutgoodhabits.com/mcp
2. Build
Tell your agent what you want. A first prompt from one of our families:
"Ben has a fractions test Friday and he keeps mixing up 1/4 and 1/8. Build him a ten-minute practice drill and pay 15 gems."
The agent authors an interactive canvas (a self-contained HTML activity), uploads it through the MCP server, wraps it in a skill, and schedules it as a task. Sprout checks the canvas before it ever reaches the kid: no external calls, no trackers, nothing leaves the sandbox.
3. The kid side
The task shows up on Ben's iPad inside Sprout, next to his other plans. He does the drill after school. He gets two wrong, the canvas shows him why an eighth is smaller than a quarter, he retries, he finishes. Sprout pays out the 15 gems.
4. The receipt
Your agent gets the result back through the same MCP connection: completed, score, time spent. Friday night you can ask your agent how the week of fraction drills went and it answers from data, not vibes.
What Sprout is scoped to
Sprout deliberately stays on the kid side of the line. Your agent owns the integrations, the reasoning, and your family's context. Sprout owns the iPad runtime, the safety review, the gem economy, and the parent approvals. The docs walk through the full protocol, and the skills marketplace has working examples you can adopt and remix.