MCP CLI

MCP CLI is a command-line inspector for MCP servers. You point it at a server, and it gives you an interactive shell to call tools, read resources, and run prompts without needing to wire up a full client. It handles OAuth flows, which matters if you're testing servers that authenticate against external services. I reach for this when I'm building or debugging an MCP server and want to confirm a tool actually works before I add it to Claude Desktop. The workflow is simple: run `npx @wong2/mcp-cli path/to/server`, pick a tool from the list, fill in the arguments, and see the raw response. It's faster than restarting Claude Desktop every time you tweak a function signature. The interactive prompt is clear, and it shows you the full JSON payloads, which helps when you're trying to figure out why a tool isn't behaving. The OAuth support is the standout feature. Most CLI tools for MCP skip this, but mcp-cli will open a browser window, handle the callback, and pass the token through. This means you can test servers that integrate with GitHub, Slack, or any other OAuth provider without mocking credentials. It's also useful for CI pipelines. You can script it to call a specific tool and check the output, which gives you a basic smoke test before you ship a server update. The exit codes are sensible, so it fits into existing test harnesses without much fuss. Quirks: it's a Node package, so you need a recent version of Node installed. The interactive mode is great for exploration, but if you want to automate a sequence of calls, you'll need to script it yourself. There's no built-in way to chain tool calls or save sessions. The output is JSON-heavy, which is accurate but not always easy to scan if you're just trying to spot an error. Skip this if you're only using MCP servers, not building them. It's a developer tool, not an end-user utility. If you're happy with the Claude Desktop interface and never need to test servers in isolation, you won't miss it.