[azf20]

amp query had http://localhost:1602 hardcoded, so the CLI couldn't query staging or any other remote server. It also exited 1 with no output on any failure.

• --url / -u falls back to AMP_QUERY_URL, then to http://localhost:1602. It accepts grpc:// and grpc+tls:// (as used by ampd, ampsql and amp-python) as well as http(s)://.
• --token now falls back to AMP_AUTH_TOKEN, then to the cached login as before.
• Failures print a reason, e.g. Query failed: [unauthenticated] Jwt is missing or Query failed: [unavailable] connect ECONNREFUSED 127.0.0.1:1602.
• A root pnpm amp script runs the CLI from source with --experimental-transform-types. The generated protobuf enums need that flag. Building doesn't help, because inside the workspace @edgeandnode/amp still resolves to src/*.ts.
• The README has a getting-started section: the CLI against staging, and SDK examples for Flight and the Admin API. It notes that @edgeandnode/amp on npm (0.0.55) is the older SDK from edgeandnode/amp.

Not changed: auth.amp.thegraph.com (used by amp auth login) serves an expired certificate, and I couldn't find a replacement host.

Testing

• pnpm check:tsgo, pnpm lint and pnpm circular pass; pnpm test: 233 passed.
• Queried staging with flags and with env vars, on Node 24.21 and 22.23. With empty env vars it falls back to the defaults; I checked the Authorization header against a local stub server.
• Ran the README's TypeScript example against staging, with and without AMP_AUTH_TOKEN.

🤖 Generated with Claude Code