Add mcp: Chamber

[abm9111]

Resubmission of #5778, closed for missing frontmatter. Thank you for the review — the diagnosis was exact and saved me guessing.

The old file was the website preflight form output pasted as markdown headings, so classify-pr read it as an added mcp entry and the schema then had nothing to validate. This one has real YAML frontmatter, with the field layout taken from content/mcp/firecrawl-mcp-server.mdx. All four contentRequired and all four contentRecommended fields are present.

Your four corrections, all applied:

1. The limitations count is 19, not 17. You were right; I counted the ## sections myself rather than trusting the old number.

2. The three install forms now agree. All of installCommand, copySnippet and configSnippet are npx -y @bu7umaid/chamber mcp. You were right that try was wrong — it builds and deletes a throwaway workspace and never starts a server; it belongs in the description as a demo, not as the install. The absolute-node form in the old configSnippet was the clone path, which does not apply to a package install, so it is gone.

I verified the form rather than assuming it: initialize and tools/list over stdio against the published tarball from a neutral cwd, returning serverInfo and the three tools.

3. prerequisites now leads with Node 23.6+, which is the floor engines declares.

4. It points at a version that works. Your last point turned out to matter more than the version number. 0.1.3 shipped with ask answering from a canned stub, unlabelled, whenever a config omitted model.mode — text that reads exactly like a real model declining for lack of sources, printed above a per-claim citation verdict block. Listing the server would have sent people straight into that on their first tool call, so I fixed it and published 0.1.4 before resubmitting. A stub answer now labels itself on both the CLI and inside the MCP tool result. It is entry 19 in KNOWN_LIMITATIONS.md.

0.1.4 also closes a shell-injection hole in the repository's own composite action, drops root in the published image, and pins its workflow actions by digest.

On the safety notes you quoted: they are in safetyNotes and privacyNotes now, with the count corrected. I kept the one about ingest having no default exclude list, since pointing it at the wrong folder is the most likely way someone gets hurt by this tool.

No existing entry references Chamber — checked across content/ for the name and the package.