Website feedback shaped for coding agents

The missing input for agent workflows

Coding agents work from text and structured context. Vague screenshots in chat lose route, selector, viewport, and failing requests. Shiproom preserves that evidence at capture time.

Two handoff paths

Export: shiproom feedback export produces deterministic Markdown by default, or JSON with --format json.

MCP: @shiproom/mcp is published on npm. Tools list projects, list and fetch feedback, and update feedback status. They operate on Shiproom feedback data, not arbitrary local source code.

The complete human-to-agent loop

The agent builds and starts the application. A person reviews the actual interface and records a visual judgment. Shiproom transports that judgment with captured context. The same agent finds the owning source, implements the smallest coherent change, and runs the repository checks. A person reviews the UI again and decides whether the outcome is right.

That final review is not ceremony. An automated test can prove the route still loads and the button still works; it cannot decide whether the new hierarchy expresses the product intent the reviewer meant.

What useful agent feedback looks like

Instead of ‘make this card better,’ give the route and state, describe what is visibly wrong, state the intended outcome, name what must remain unchanged, and define what the agent should verify. The selected element, screenshot, viewport, and nearby errors are evidence alongside that human direction—not a substitute for it.

For a reusable six-part packet, use the visual-feedback guide. For implementation, attach the export to the UI review prompt template. These resources are deliberately separate from the setup prompt that opens a review session.

Best fit and wrong fit

Shiproom fits a builder using an agent who needs visual judgment from a client, founder, designer, QA partner, or teammate before deployment. It also fits a solo builder who wants a durable contextual handoff rather than a screenshot pasted into chat.

A native agent design mode or local extension is simpler when the builder alone is reviewing on the same machine. A broader feedback suite is better for permanent production widgets, formal approvals, large QA queues, surveys, or session replay.

Which agents this has actually been tested with

Codex CLI is directly verified end to end against the published adapter: token creation, authentication, feedback retrieval, and token revocation. Cursor, Claude Code, and other clients that speak the same stdio MCP contract can use the documented configuration, but Shiproom does not label them directly verified without a recorded end-to-end run.

This is compatibility, not a commercial relationship. Shiproom has no partnership, endorsement, or certification from Anthropic, OpenAI, or Anysphere, and does not appear in any of their official directories. All of those tools can read a pasted Markdown or JSON export without MCP.

Boundaries

Shiproom does not read your repository, does not write into your project directory, and does not claim to fix issues autonomously. It prepares the task; a human or an agent does the work.

Keep reading

• Shiproom vs Webvizio
• Shiproom vs Userback
• Shiproom vs BugHerd
• Localhost review
• Visual feedback tools
• Free website feedback
• Every comparison we have researched
• What Shiproom costs today