Getting started with local MCP server, how?
retained text
EDIT: I was able to run this in Pi. This whole confusion came to my mind because it is called **server** when for my point of view is more like a **script** what is run my harness when needed and that script have just "public methods" for coder's point of view. Eg. tools defined on that script are just methods/functions what are able to call by harness/agent.
Anyway here is the sample config for my Pi in \~/.pi/agent/mcp.json and it automatically detected all methods there.
{
"mcpServers": {
"my-server": {
"command": "/home/user/mcp-server/.venv/bin/python",
"args": [
"/home/user/mcp-server/server.py"
],
"exposure": "direct"
}
}
I am trying to learn MCP servers. I used AI to give me boilerplate code what should just return text to me depending on what MCP tool I call. It is python server and it uses MCPServer package.
Anyway, I have my server up and running, but now what? How I can test it with most easy way possible with my local AI?
I tried first llama-ui to connect to it, but then I found out that it is stdio by default and not http and there is no easy possibility to connect to that server from webui without doing more configurations with --webui-mcp-proxy and other stuff. I thought that well, not worth spending time with that since all what I just want to is test if that MCP server works and if local AI agent tools can connect to it and use those tools. This started to get off the rails too far so then I moved to LM Studio.
On LM Studio, when I want to create local MCP server, I need to configure my local MCP server in JSON and manually tell all the tools it support, if I understand correctly? So if I edit my MCP server, add new functionality, I need again to update my MCP configuration manually so it starts to lose its whole point, at least on this testings. Have I misunderstood that how LM Studio configs work on these?
Also on LM Studio config JSON I need to give script path? I thought that MCP server is a server what is up-and-running and my clients like LM Studio can connect to that server and ask what tools it provide and then it can use those without I need to define "here are apps you can run on that path"?
So, HOW I can use my local MCP Server what works over stdio to even test that functionality? What harness I should use to test this as easily as possible to get started?document fetched 2026-10-24 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xd9hc3/getting_started_with_local_mcp_server_how