so, uh, MCP Inspector?
Closed this issue · 1 comments
How do I connect to the mcp-server-metatool server via MCP Inspector? I am trying to test the server to make sure the tools that I'm adding are showing up, before I add a ton of them.
So, Inspector asks for:
- Transport Type (STDIO, SSE)
- Command (I'm assuming it's uvx here)
- arguments (so, in this case, normally in your config it would be
mcp-server-metatoolbut since it's running in Docker, I'm not sure that's going to be sufficient...? - Environment variables (I've added the API_KEY and _BASE_URL as variables)
All I'm getting is a connection error. No indication of why it's failing. Not sure how to connect to it, basically.
I configured it as per the instructions in the git repo. The server is running. Here is what I have configured:
"metatool": {
"command": "/Users/<<username>>/.local/share/mise/installs/python/3.13.1/bin/uvx",
"args": ["mcp-server-metatool"],
"env": {
"METATOOL_API_KEY": "<<really long api key from the api keys page in the GUI>>",
"METATOOL_API_BASE_URL": "http://localhost:12005"
}
}Good question, tbh I currently do see some tools are having issues. I initially add server names to route tool calls by namespace, but some server names contain "/" and cannot pass the regex of a tool in MCP. So I updated this sdk to 0.0.2 to sanitize the server name. I agree that using MCP inspector to debug is a good way and appreciate the effort there.
Here is how I use inspector (if tools are compatible):
In bash use
npx @modelcontextprotocol/inspector uvx mcp-server-metatoolThen in localhost:5173 web UI
Make sure METATOOL_API_KEY is added in the UI (see screenshot). I tried export that env var in bash and start the inspector, but that doesn't work.
I see and understand your pain there but unfortunately I also don't know why connecting to the raw py bin or file raise such errors.. but at least from using my command it should boot up the inspector in my case.
