dev-proxy: hardcoded 120s build timeout makes docker rebuilds report a successful build as 'Build failed'

[debugmcpdev]

Summary

dev_rebuild_and_restart hardcodes a 120 s build timeout. A cold-cache npm run docker-build routinely exceeds that (the image is ~1.5 GB), and when the timeout trips the proxy reports Build failed: followed by a truncated tail of the Docker log — even though the build itself is fine. Re-running npm run docker-build by hand immediately afterwards succeeds and produces the image.

So the tool reports a successful build as a failure, and the accompanying output is a mid-RUN Dockerfile step rather than any diagnosis.

Current behavior

// tools/dev-proxy/dev-proxy.mjs:266
result = execSync(BUILD_CMD, {
  cwd: PROJECT_ROOT,
  encoding: 'utf-8',
  stdio: ['ignore', 'pipe', 'pipe'],
  timeout: 120000,          // hardcoded
  env: { ...process.env },
});
} catch (err) {
  const output = [err.stdout, err.stderr].filter(Boolean).join('\n') || err.message || String(err);
  throw new Error(`Build failed: ${sanitizeStderrTail(output, { maxLines: 20, maxChars: 2000 })}`);
}

execSync surfaces a timeout kill as a thrown error just like a non-zero exit, and the catch treats both identically. err.killed / err.signal === 'SIGTERM' are available and unused.

Observed

dev_rebuild_and_restart  (mcp-debugger-docker)
→ { "success": false,
    "error": "Build failed: …cp -r /app/packages/adapter-rust/dist
               /app/node_modules/@debugmcp/adapter-rust/ (last 20 of 1454 lines)" }

$ npm run docker-build      # same command, run directly
… #69 naming to docker.io/library/mcp-debugger:local done
$ echo $?
0

Note the 1454-line build log truncated to its last 20 lines — for a timeout that tail is always an arbitrary mid-build step, never the reason.

Requests

1. Make the timeout configurable, alongside the other knobs the proxy already exposes as env vars (DEV_PROXY_PORT, DEV_PROXY_BUILD_CMD, DEV_PROXY_ROOT, DEV_PROXY_BACKEND_TRANSPORT) — e.g. DEV_PROXY_BUILD_TIMEOUT_MS. A docker-group proxy could then be configured with a realistic ceiling.
2. Raise the default, or make it depend on the configured BUILD_CMD. 120 s is comfortable for npm run build but not for an image build.
3. Distinguish a timeout from a build failure. On err.killed/SIGTERM, say so — "Build timed out after 120000 ms (raise DEV_PROXY_BUILD_TIMEOUT_MS); the build may still have succeeded, re-run manually to confirm" — instead of Build failed:. This is the part that actually costs time: the message sends you looking for a build regression that isn't there.

Impact

Dev tooling only (tools/dev-proxy/), no effect on the shipped server. Worth fixing because the docker group is unusable through dev_rebuild_and_restart on a cold cache, and the misreport is convincing enough to derail a debugging session — it did during a /testdebugger sweep, where the docker group had to be rebuilt out-of-band and restarted with dev_restart_debugger.