MCP read report: pages without a description

An agent's report from one filtered list_pages call: which /docs pages lack a description, and what that proves.

A report from an agent, written from one list_pages call through the goatboat MCP server against local goatboat:

json
{
  "under": "/docs",
  "empty": ["description"],
  "include": ["excerpt", "file", "targetUrl"]
}

One call returned exactly the pages that need work, with enough context to write each description. No page had to be opened.

What came back

8 pages under /docs have no description. Across the whole site, 27 of 126 pages are missing one.

Code pages (no body, so no excerpt)

PathStatusFile
/docs/body-textpublishedsrc/app/docs/body-text/page.tsx
/docs/link-hoverspublishedsrc/app/docs/link-hovers/page.tsx
/docs/muted-textpublishedsrc/app/docs/muted-text/page.tsx
/docs/text-colorspublishedsrc/app/docs/text-colors/page.tsx
/docs/id-flowdraftsrc/app/docs/id-flow/page.tsx

To describe these, an agent reads the page.tsx named in file.

/docs/id-flow is the test page deleted earlier. Its row is flagged missing, but the tree still lists it, so listings should probably skip missing rows too.

Markdown pages (excerpt included)

PathStatusOpens with
/docs/mcp-wire-formatdraft"When an agent creates a page through the goatboat MCP server, it does not send a markdown file…"
/docs/page-identitydraft"Goatboat keeps a row for every code page… matches them to rows by path."
/docs/test-docdraft"This is a throwaway page for testing the goatboat MCP proposal flow."

For these, the excerpt alone is enough to write a one-line description.

What this proves

  • Filtering works: pages that already have a description, such as /docs and /docs/heading-fonts, were left out.
  • Field selection works: only the requested extras came back, and each is null where it doesn't apply.
  • The loop is complete: read with list_pages, then write every description in one propose_changes batch for a single approval.

Found along the way

  • Two goatboat MCP servers were registered, one in the VS Code user config and one in the repo. That's why the tools were named goatboat2, and why the first attempts reached production instead of local goatboat.
  • What the MCP create_page call actually sends

    What a goatboat MCP create_page call actually sends, hop by hop from the agent's arguments to the proposal request.

  • Proposing complete, published pages

    Agents can now propose a new page with its description, tags, project and section, published on approval. What changed, and the security questions that come with it.

  • Content tree

    Every node nested under its parent. Paths, and so URLs, come from this tree.

  • Tags

    Every tag with its count. Each links to the feed filtered by that tag.