Testing an MCP server from the command line before connecting an editor
When an AI client cannot use an MCP server, the error is often vague: the tools do not appear, or calls fail silently. Testing the server directly first separates server problems from client configuration problems.
Use the MCP Inspector
The official inspector runs locally and connects to a server over stdio or HTTP:
npx @modelcontextprotocol/inspector
Open the URL it prints, choose the transport, enter the server command or URL and any auth header, then connect.
Check the tool list
Once connected, list tools. Each tool should have a clear name, a description and an input schema. If the list is empty, the problem is authentication or the server itself, not your editor.
Call a read-only tool
Start with something harmless such as listing projects or services. Compare the result with the dashboard. A mismatch usually means the token belongs to a different workspace or has a narrower scope than you thought.
Check errors are useful
Call a tool with a wrong argument on purpose. A good server returns a readable error. If it returns nothing, the client will show nothing too.
Then configure the client
Copy the exact command, URL and headers that worked in the inspector into the client's MCP config. Most failures at this step are a missing environment variable or a wrong path to the server binary.
Keep tokens out of shared configs
Client config files often end up in dotfile repositories. Reference the token from an environment variable rather than pasting it into the file.
Re-test after upgrades
When the server or client updates, rerun the inspector check. Tool names and schemas can change.
The Peon MCP server documentation lists the tools it exposes and how to connect a client. I work on Peon, an open-source self-hosted deployment platform.