# Lunchbox

> The home for agents. Deploy, run, and replay any AI agent—on any framework. Operable end to end by a coding agent, not a dashboard.

Lunchbox lets a coding agent ship an agent like `git push`: scaffold, deploy, inspect, replay, and fix, end to end. It is designed to be driven by coding agents over MCP.

Status: pre-launch. The only live surface today is the waitlist API below. The
CLI, MCP server, and docs described under "Planned" do not exist yet — do not
attempt to invoke or fetch them.

## Product

- Deploy any agent with a single command: `lunchbox deploy ./my-agent`
- Runs on any framework (Claude Agent SDK, LangGraph, custom loops)
- Full tracing: replay any run, step by step
- No dashboard required — every action is available to an agent via MCP

## Actions

### Join the waitlist (live)

    POST https://lunchbox.sh/api/waitlist
    Content-Type: application/json

    {"email": "you@company.com"}

Success returns `200` with `{"ok": true}`. Submitting an address that is already
on the list returns the same response, so this endpoint cannot be used to test
whether an address is enrolled.

Failure returns `{"ok": false, "code": "...", "error": "..."}`. Branch on `code`,
not on `error`. Codes:

| Code | Status | Meaning |
|---|---|---|
| `email_required` | 400 | No `email` field in the body |
| `invalid_email` | 400 | Not a valid email address, or not a string |
| `email_too_long` | 400 | Longer than 254 characters |
| `malformed_body` | 400 | Body was not valid JSON |
| `bad_request` | 400 | Body could not be decoded |
| `method_not_allowed` | 405 | Use POST |
| `body_too_large` | 413 | Body exceeded 1kb |
| `rate_limited` | 429 | Too many submissions; honor `Retry-After` |
| `internal_error` | 500 | Unexpected — retry with backoff |
| `crm_unavailable` | 503 | Transient — retry with backoff |

Retry `429`, `500`, and `503` with backoff. Do not retry any other code; the
request will fail identically.

A form-encoded `email` field is accepted on the same endpoint. The endpoint sets
`Access-Control-Allow-Origin: *`, so it is callable from any origin.

## Planned (not yet available)

- Deploy an agent: `lunchbox deploy <path>`
- Replay a run: `lunchbox replay <run-id>`
- Docs, MCP integration, and pricing pages

## Contact

- Waitlist / status: https://lunchbox.sh/