How can Inspector distinguish transport, discovery and authorization failures?
MCP Inspector is a reference developer tool with web, CLI and TUI workflows. It is useful for separating a client's presentation from the underlying remote tool call, but its own local proxy credentials must remain private. This guide assumes Inspector is already installed and uses its documented remote HTTP flags. It neither installs a package nor treats a successful transport handshake as authorized Montage access.
--server-url https://app.montages.co/mcp --transport http--server-url https://app.montages.co/mcp --transport http --cli --method tools/call --tool-name signals_statusCheck the installed Inspector interface
Find the existing Inspector executable or approved project command and inspect its documented help for the installed version. Do not use an automatic package runner that downloads dependencies. Select the web, CLI or TUI workflow appropriate to your environment, and keep the local proxy's access controls intact rather than disabling protection to simplify a screenshot.
Target the remote HTTP server
Use --server-url https://app.montages.co/mcp with --transport http in the documented remote workflow. These flags select the hosted server instead of launching a stdio subprocess. Check the complete endpoint before testing; changing the transport to an SSE example or pointing at the marketing site will not diagnose the same connection.
Handle auth and local secrets separately
Configure supported Montage authorization for the Inspector setup you are using. OAuth handling and local secret storage depend on that setup, so verify the documented path rather than assuming a browser session is inherited. Keep Inspector proxy tokens, authorization material and consent details out of screenshots, terminal transcripts and chat. Do not weaken local proxy security to bypass remote auth errors.
Inspect tool metadata first
Once the authorized session is available, inspect the tools interface and locate signals_status. Note whether a failure occurred during initialization, tools discovery or tools execution; those are different protocol stages. Tool metadata can help determine the expected input, but it is not an execution receipt and should not be represented as a workspace status response.
Run the read-only signals_status check
Use the documented CLI method tools/call with --tool-name signals_status, or select the same tool in the installed Inspector interface. Inspect the actual workspace response. If authorization fails, repair the supported auth setup; if HTTP initialization fails, revisit flags and endpoint. No paid model call, collection or write is required. Keep setup blocked if the real authorized tool call cannot complete.
Keep the boundaries clear.
- Use an installed Inspector; no automatic dependency download is part of this guide.
- OAuth and secret storage are setup-dependent, and local proxy tokens must stay private.
- Inspector diagnostics are not a live certification of every downstream MCP client.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-13. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.