You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The MCP server can write valid YAML (write_page) and show pre-canned component screenshots (preview_component), but an agent cannot see the rendered result of the content it actually wrote. The agent composes blind — it knows the schema is valid but can't verify visual correctness.
This is the gap between "schema-valid" and "looks right."
Why it matters
This is the single highest-leverage feature for a hackathon demo. It turns the story from "trust me, the schema validated" into "look at what the AI just built." It completes the AI authoring loop: introspect schema → write content → see rendered result → iterate.
Problem
The MCP server can write valid YAML (
write_page) and show pre-canned component screenshots (preview_component), but an agent cannot see the rendered result of the content it actually wrote. The agent composes blind — it knows the schema is valid but can't verify visual correctness.This is the gap between "schema-valid" and "looks right."
Why it matters
This is the single highest-leverage feature for a hackathon demo. It turns the story from "trust me, the schema validated" into "look at what the AI just built." It completes the AI authoring loop: introspect schema → write content → see rendered result → iterate.
Depends on
Proposed approach
Add a
stackwright_preview_pageMCP tool that:slugparameter (page must already exist viawrite_pageoradd_page)The infrastructure is 90% there:
packages/e2e/@stackwright/build-scriptsOpen questions
pnpm devrunning?Acceptance criteria
write_pagethenpreview_pageand receives a screenshot of the rendered page