Integration proposal: Tupper (AI agent sandboxes, MCP server) (auto-drafted 2026-06-29)

[NativeTeachingAidsB]

Score: 7 — new MCP-native sandbox launched June 23; early integration positions codetalker as first-mover. Labels to add manually: integration-proposal, auto-generated.

Auto-generated by the weekly-landscape-scan routine. Draft spec in docs/superpowers/specs/2026-06-29-integrate-tupper.md (on branch docs/landscape-report-2026-06-29). Do not promote to approved without human review.

What is it?

Tupper (MIT, launched June 23, 2026, 137 stars in <1 week) — open-source sandboxes for AI agents. Runs untrusted AI-generated code safely via Apple Containers (macOS), Firecracker (Linux), WSL (Windows). Ships @tupper/mcp — a MCP server exposing sandbox lifecycle.

Why it matters

As agents run generated code in sandboxes, narrating sandbox lifecycle (boot → exec → success/error) is a natural extension of codetalker's live cadence mode. Since both are stdio MCP servers, co-loading them in Claude Code or Codex requires zero glue code.

Integration

{
  "mcpServers": {
    "codetalker": { "command": "codetalker-mcp", "transport": "stdio" },
    "tupper":     { "command": "npx", "args": ["@tupper/mcp"], "transport": "stdio" }
  }
}

Key open questions

• Which lifecycle events does @tupper/mcp expose? (Needs source review.)
• Is Firecracker (Linux) path stable yet?
• Should we wait for Tupper v1.0 stable API before publishing the guide?

Effort

Small — 1–2 days (integration note + validation). Mark as experimental pending Tupper v1.0.

---

Routine run: 2026-06-29 | Landscape report: docs/landscape-reports/2026-06-29.md