Feature flags, A/B experiments, translations, and a feedback→fix loop — operated end-to-end by your coding agent. Describe what you want; it writes the code, ships it, and watches the metrics. Free to start, installs in one command.
Scroll. The panel pins, the demo morphs.
{
"max_uploads": 100,
"model": "opus-4.1",
"enable_v2": true,
"region": "us-east-1"
}Every flag is a killswitch. Pull it manually, or set a guardrail — error rate, p95 latency, spend — and Shipeasy yanks it for you.
Typed runtime configs targeted by user, plan, geo, or anything you send. Edit from Claude, the dashboard, or a PR.
Sequential stats mean you can peek whenever you want without inflating false positives. Auto-ramping, SRM detection, guardrails — Claude writes the wrapper.
Activation, retention, revenue, and error rate are instrumented from day one. Add custom metrics with one line of code.
The feedback widget catches it, Shipeasy triages it, your agent fixes it.
The in-app feedback widget and the see() error primitive turn a bug or feature request into a structured, deduplicated item in your dashboard.
Each report lands with the failing code path, the consequence chain, and a priority — ready for an agent to act on, not a screenshot to decipher.
A scheduled Claude Code routine pulls the open queue, writes the fix on its own branch, and opens a pull request that closes the issue.
Ship the fix behind a gate, ramp it on a guardrail, and watch the error rate fall on the auto-collected metrics — all in the same platform.
One command — claude mcp add shipeasy — registers four tools with Claude. We use your project context to wire experiments into the right files. No YAML.
No. Shipeasy is an MCP server, so the core — flags, experiments, configs, and i18n — works in any MCP-capable agent (Claude Code, Cursor, Windsurf, and others) by talking to the agent you already use. The deepest automation — the slash-command skills and the scheduled feedback→fix→PR routine — runs on Claude Code today, with more harnesses on the roadmap. You're never locked into a tool you don't already use.
Both. Hosted is the default. Self-hosted is a one-binary deployment on Enterprise with full parity and an air-gapped mode.
Sequential testing by default — peek whenever, without inflating false positives. Bayesian and frequentist are available per-experiment.
Every action is reversible and logged. Killswitches and experiment changes go through a PR or a tool call you can undo.
JS/TS, Python, Go, Ruby, Rust, Swift. The SDK is a thin client; the heavy lifting happens server-side.
Install in 12 seconds. Your first experiment before your coffee is cold.