Give your coding agent visual feedback on localhost.

No. The usual way to start a review is to paste one prepared prompt into the coding agent you are already building with, such as Claude Code or Codex, and let it do the setup. It starts your app if it is not running, opens Shiproom, walks you through a one-time browser sign-in, waits until the review session is live, and replies with the private reviewer link. The terminal command is documented on this page for people who prefer it, but it is an alternative route, not a requirement.

One prepared prompt, which you copy from the top of this page with the Copy the review prompt button. It tells your coding agent to start the app if it is not already running, run Shiproom in the project folder, walk you through the one-time browser sign-in, wait until the review session is live, reply with the private reviewer link, keep the session running until you say stop, and leave your code alone while it is open.

No. Reviewers open the link in an ordinary browser. There is no extension, no desktop app, no download, and no account to create when you invite them through an invite link. They point, draw, select an element and write a comment directly on the running interface.

Yes, by default. A project is invite-only when it is created, so only someone holding an invite link can open it. Invites can be listed, rotated and revoked at any time, and revoking ends that link immediately. Treat an invite link as access-bearing and share it only with the people you intend to review the work. You can switch a project to public, and the CLI asks you to confirm before it does.

No. Shiproom never reads your source code, your Git history or metadata, or your package manifest. It does not inspect your repository or your project contents, and it does not use package.json to identify your project; a project is resolved from the folder path alone. Reviewer traffic is forwarded to your running server over HTTP, and that is the only part of your application Shiproom sees.

No. Your application keeps running where it already runs, on your own machine. Shiproom creates a private link and forwards reviewer traffic to that running application for the duration of the session. It is not a deployment platform, it does not replace your hosting, and the reviewer link is not a permanent address.

The reviewer link stops serving your application and reviewers see an offline page instead. A session is offline if the local app stops, the CLI stops, the computer sleeps, or you lose internet access. Feedback already submitted stays in your project and remains readable in the console and the CLI, so nothing your reviewer sent is lost.

Yes. Export a report as Markdown or JSON and give it to any coding agent that reads text, or connect a compatible agent through the published @shiproom/mcp adapter so it can inspect Shiproom feedback packets directly. Shiproom prepares the evidence; a person or an agent makes the change. Shiproom does not edit your repository and does not claim to fix issues autonomously. Codex CLI is the directly verified MCP client. Cursor, Claude Code, and other stdio-compatible clients can use the same adapter configuration, but Shiproom does not publish them as directly verified without a recorded end-to-end run.

No. Shiproom has no partnership, affiliation, endorsement or certification from Anthropic, OpenAI, Anysphere or any other vendor, and it is not listed in any vendor's official directory. Markdown and JSON exports are portable text, and the MCP adapter speaks the public stdio protocol. Codex CLI is the client directly verified end to end. Compatibility is not a commercial relationship.

Yes. Shiproom is currently free to use. There is no paid plan and no trial clock. Early-access per-account limits currently apply: 3 projects, 100 stored feedback items, and 30 days of feedback history. Reviewers invited by link do not need a Shiproom account or installation. See the pricing page for the full plain-language summary.

Shiproom turns a web app running on your own machine into a private reviewer link. You get the link by asking your coding agent for it, or by running one command yourself. Whoever you send it to opens the app in an ordinary browser and leaves visual feedback directly on the live interface. Shiproom captures the contextual evidence around each comment and exports it so you can hand that feedback to a developer or a coding agent.

When a reviewer submits feedback, the report can carry the route and page title, a screenshot, the geometry of each annotation, metadata about the selected element such as its tag, id, accessible name and a stable selector where one is available, the viewport, the browser and operating system, the timestamp, the reviewer label, the chosen category and the comment, along with failed HTTP requests and JavaScript errors recorded shortly before submission. Not every field is present on every report; what is available is what was captured at that moment.

Start your application as you normally would, then run npx shiproom@latest live in the same folder. Shiproom signs you in on first use, finds the local HTTP server, resolves the project for that folder, and prints the private reviewer link. Use --port when you want to name the port explicitly. This is exactly what the review prompt asks a coding agent to do on your behalf.

No. Shiproom writes no files into your application directory: no config file, no lockfile entry, no directory of its own. Local session state lives only in Shiproom's own configuration directory elsewhere on your machine. It does not change your code automatically, and starting a review changes nothing about how you build.

Yes. Run shiproom feedback export with a reference to get deterministic Markdown on standard output, or add --format json for JSON, or --out to write a file. The same report always produces the same bytes, with object keys sorted. Shiproom does not use a language model to generate exports.

Yes, as an optional extra rather than the main path. @shiproom/mcp is published on npm. Create a personal token with the Shiproom CLI, then run the adapter with npx -y @shiproom/mcp using SHIPROOM_TOKEN and SHIPROOM_API_URL. It exposes tools to list projects, list and fetch feedback, and update feedback status. Those tools operate on Shiproom feedback and project data, not on arbitrary local source code. The recorded production end-to-end verification covered token creation, authentication, feedback retrieval, and token revocation through Codex CLI.

The current published release is shiproom 0.4.3, available on npm. Run it with npx shiproom@latest live, or install it globally with npm install --global shiproom. It requires Node.js 20 or later and runs on macOS, Linux and Windows.

The CLI runs anywhere Node.js 20 or later runs, which covers macOS, Linux and Windows. On the framework side there is nothing to support individually: Shiproom forwards requests to whatever HTTP server you are already running on a local port and injects the review widget into HTML responses, so it is not tied to a particular framework, bundler or build tool.

Yes. The CLI is published on npm and accounts are open to sign up. It is still early, so per-account limits apply while capacity is managed: 3 projects, 100 stored feedback items and 30 days of feedback history. Those limits are visible in the console and will change as the product matures. Shiproom is not intended for large organisations that need formal compliance assurances.