MCP Inspector CLI: list tools and call one from the terminal

MCP Inspector is the MCP project's own test client. Most people know its web UI, but it also has a CLI mode: add --cli, name one method, and it prints the result and exits. That makes it the fastest way to answer one question when a client shows a red dot or "failed": is the server broken, or is my client config?

The commands below were checked against the Inspector README as of this post's date. Flags change between versions, so run npx @modelcontextprotocol/inspector --help if one is rejected.

List the tools of a local server

For a server your client starts over stdio, give Inspector the same command and arguments:

npx @modelcontextprotocol/inspector --cli uvx mcp-server-fetch \
  --method tools/list

Inspector starts the process, runs the MCP handshake and prints the tool list as JSON. If this fails, the command itself is wrong or the runtime is missing, and no client config will fix it.

List the tools of a remote server

For a Streamable HTTP server, pass the URL, the transport and the header your client sends:

npx @modelcontextprotocol/inspector --cli https://fetchmcp.net/mcp \
  --transport http --method tools/list \
  --header "Authorization: Bearer $LIVE_WEB_DATA_API_KEY"

Use the exact URL from your client config, including the path. A server that answers here but not in your client is fine; the problem is in the client entry, often the transport type or a variable that was never expanded.

Call one tool

Listing tools proves the connection. Calling one proves the key and the account. Pick a tool that does no real work:

npx @modelcontextprotocol/inspector --cli https://fetchmcp.net/mcp \
  --transport http --header "Authorization: Bearer $LIVE_WEB_DATA_API_KEY" \
  --method tools/call --tool-name get_account_status

Arguments go in as --tool-arg key=value, one flag per argument, for example --tool-arg url=https://example.com for a fetch tool. Start with the defaults before you add size or format options.

What each result tells you

The full order of checks, from curl to client logs, is in the debugging runbook.

Try it against a real server

Live Web Data API, made by Teneo Labs, runs a remote Streamable HTTP server whose get_account_status tool costs nothing, so the tool call above is a free smoke test once you have a key. The current URL and client configs are in the MCP docs.