feat(mcp): add --host flag (and QMD_HOST env) to the HTTP server

[djalmajr]

Problem

startMcpHttpServer hardcodes httpServer.listen(port, "localhost", …), so the HTTP MCP server only accepts loopback connections. When qmd mcp --http runs inside a container, a Kubernetes liveness/readiness probe connects from the pod IP (e.g. 10.42.x.x), not 127.0.0.1 — the bind refuses it and the probe fails, so the pod never goes ready.

Today the only workaround is to patch the published dist/mcp/server.js at image-build time:

sed -i 's@listen(port, "localhost"@listen(port, process.env.QMD_HOST || "localhost"@g' "$QMD_DIST"

which is brittle and breaks on every release.

Change

Add a --host flag to qmd mcp --http, with a QMD_HOST environment-variable fallback:

host = options.host ?? process.env.QMD_HOST ?? "localhost"

• Default is unchanged (localhost) — no behavior change out of the box.
• --host 0.0.0.0 (or QMD_HOST=0.0.0.0) binds all interfaces for container/probe reachability.
• The flag is forwarded to the detached process in --daemon mode and reflected in the startup log / Started on … message.

Files:

• src/mcp/server.ts — startMcpHttpServer resolves and binds host; JSDoc + listening log updated.
• src/cli/qmd.ts — --host flag parsed, forwarded to daemon spawn, shown in logs.
• README.md — documents --host / QMD_HOST.

Verification

• npm run test:types — passes.
• test/mcp.test.ts — 57/57 pass (existing HTTP test uses the default and still binds localhost).
• Functional smoke:

  $ qmd mcp --http --host 0.0.0.0 --port 18234
  QMD MCP server listening on http://0.0.0.0:18234/mcp
  $ lsof -nP -iTCP:18234 -sTCP:LISTEN
  node … TCP *:18234 (LISTEN)        # * = all interfaces
  $ curl -s -o /dev/null -w '%{http_code}' http://127.0.0.1:18234/health
  200