Inspect a Runtime Profile with MCP Inspector

Use MCP Inspector to separate a Runtime Profile or server problem from an AI client problem. LightNow generates a small executable wrapper; Inspector starts that wrapper as a stdio MCP server and exercises the same runtime context used by lightnow run.

What this test covers

Boundary	What success proves
Profile export	The profile contains a Registry-linked, runner-compatible server.
Runtime readiness	Required config and secrets resolve for the selected context.
Server startup	The command, package, environment, and working directory are usable.
MCP discovery	The server answers a real tools/list request.

This is a focused runner test, not LightNow Proxy mode. Day-to-day AI clients should still connect through LightNow Proxy.

1. Generate the wrapper

Create an executable Inspector wrapper

lightnow sync --client mcp-inspector --runner --profile default --config-path ./lightnow-mcp-inspector.sh

The profile must contain at least one Registry-linked server with an alias and version. The generated shell wrapper targets the first runner-compatible server. Use a focused profile when you need to test a particular upstream.

The wrapper has this shape:

lightnow-mcp-inspector.sh

#!/usr/bin/env sh
exec lightnow 'run' '--profile' 'default' '--server' 'sonarqube' "$@"

An organization selection adds the tenant argument. lightnow run requests the authorized stdio runtime context, resolves secrets for the local runner, and stops before server startup when required inputs are missing.

2. Open Inspector

Start the Inspector UI

npx @modelcontextprotocol/inspector ./lightnow-mcp-inspector.sh

Open the local URL printed by Inspector. Connect to the generated stdio server, run tools/list, inspect the tool schemas, and then call a small read-only tool.

For a repeatable command-line check, use Inspector's CLI mode:

List tools without the browser UI

npx @modelcontextprotocol/inspector --cli ./lightnow-mcp-inspector.sh --method tools/list

The command shape follows the official MCP Inspector documentation.

3. Interpret the failure boundary

Failure	Owner	Next action
Wrapper generation fails	Runtime Profile	Add a linked server with alias and version, or choose the correct profile/context.
Runner stops before startup	Runtime config	Complete the required config or secrets for that profile server.
tools/list fails	Upstream runtime	Check executable, package, environment, authentication, TLS, and server logs.
Inspector works but the AI client fails	Client or LightNow Proxy	Check client posture, proxy config, restart state, and tool cache.

Do not broaden the investigation until you know which row failed. Inspector working is strong evidence that the profile server can behave as a normal stdio MCP server; it does not prove that another client's generated config is current.

Keep Inspector local

Use the authenticated local URL printed by Inspector. Do not disable its authentication or expose its ports to an untrusted network.

Continue with Debug LightNow Proxy when Inspector works but a managed client fails. Use runtime secrets and config when the runner reports missing inputs, and runtime events for request evidence from a real client session.