Problem
Verification is done via multi-step manual round-trips (curl -I → read X-Content-Source, grep both minified+unminified, probe serve path, Playwright when curl is blocked). Each round-trip pulls output into context and costs tokens; the pattern is well-documented in agent memory but re-derived each time.
Proposal
A verification/probe MCP that takes a slug or URL and returns the compact verdict the agent would otherwise assemble by hand:
- serve-path probe (
X-Content-Source: db-current vs db, HTTP status, redirects)
- deployed asset check (grep both minified and unminified forms)
- optional Playwright fallback for Akamai-blocked / non-browser-UA paths (Akamai 403s node/axios/Claude-Code UAs)
- returns
{ ok, source, status, notes }
Impact
Removes repeated manual curl+grep sequences during "test X in DEV" / "is the fix live" checks.
Problem
Verification is done via multi-step manual round-trips (
curl -I→ readX-Content-Source, grep both minified+unminified, probe serve path, Playwright when curl is blocked). Each round-trip pulls output into context and costs tokens; the pattern is well-documented in agent memory but re-derived each time.Proposal
A verification/probe MCP that takes a slug or URL and returns the compact verdict the agent would otherwise assemble by hand:
X-Content-Source: db-currentvsdb, HTTP status, redirects){ ok, source, status, notes }Impact
Removes repeated manual curl+grep sequences during "test X in DEV" / "is the fix live" checks.