User-agent: *
Allow: /
Disallow: /api/
Disallow: /auth/
Disallow: /me
Disallow: /new
Disallow: /healthz
# /mcp is the POST-only JSON-RPC endpoint. A GET returns 405, which Search Console
# files as "Blocked due to other 4xx issue" — so keep crawlers off it entirely. MCP
# clients POST directly and never consult robots.txt, so this costs us nothing.
# /developers is the indexable page for the MCP server and stays in the sitemap.
Disallow: /mcp
# NOT disallowed, deliberately: /playbooks/*.md. Same reasoning as /r/ below — some
# agent fetchers do respect robots.txt, and fetching a playbook by URL is the
# documented fallback for MCP clients without slash commands.
# /r/ and /s/ review links are intentionally crawlable so an agent handed the
# link can fetch it; they're kept out of search by an X-Robots-Tag: noindex
# header (see _set_cache_headers), not by a Disallow. A Disallow would block the
# fetch and the agent-handoff path with it.

Sitemap: https://cobaltcapture.com/sitemap.xml