Local MCP Tools Collection

A small collection of Model Context Protocol (MCP) tools, built for local LLMs. One venv, many options.

Why it exists

Many MCP servers are distributed as separate projects and need separate setup. This tool pack keeps a few local MCP servers in one repo and one uv environment.

Features

• MCP JSON configuration generation: run main.py and go through the wizard
• One venv for multiple MCP servers

MCP Servers

• Web Search: Use duckduckgo as search engine, fetch and summarize top results
• Python SandBox: Allow agents to run Python and use NumPy and SymPy for math tasks
• Longterm-Memory: Store lightweight long-term notes

Security Notes

1. Default transport is stdio. You can switch to HTTP in GlobalConfig.
2. python-sandbox.py uses exec() and eval() for run_python. This is not a secure sandbox.
3. Treat run_python as local code execution by the agent under your user account.
4. The generated config now sets PYTHON_SANDBOX_LOG_PATH for local python-sandbox entries so every run_python execution is appended to a JSONL audit log.

Requirements

• Python >= 3.13
• Managed with uv

Install

Using uv:

uv sync

Generate LM Studio Config

Run:

uv run python main.py

LM Studio currently follows Cursor-style mcp.json notation. The generated LM Studio output uses the mcpServers object with either:

• command + args + env for local stdio servers
• url + headers for remote HTTP servers

If you select the local python-sandbox server, the wizard will ask for an audit log file path.

Run the MCP Server

python python-sandbox.py

The server communicates over stdio (FastMCP). Point your MCP-compatible client at the executable command above.

Tool Usage Examples

Run main.py for JSON configuration generation. For LM Studio, you will get something like this:

{
  "mcpServers": {
    "memory": {
      "command": "E:\\LMStudio\\mcp\\lmstudio-toolpack\\.venv\\Scripts\\python.exe",
      "args": [
        "E:\\LMStudio\\mcp\\lmstudio-toolpack\\MCPs\\Memory.py"
      ],
      "env": {}
    },
    "python-sandbox": {
      "command": "E:\\LMStudio\\mcp\\lmstudio-toolpack\\.venv\\Scripts\\python.exe",
      "args": [
        "E:\\LMStudio\\mcp\\lmstudio-toolpack\\MCPs\\python-sandbox.py"
      ],
      "env": {
        "PYTHON_SANDBOX_LOG_PATH": "E:\\LMStudio\\mcp\\lmstudio-toolpack\\data\\python-sandbox-audit.jsonl"
      }
    },
    "websearch": {
      "command": "E:\\LMStudio\\mcp\\lmstudio-toolpack\\.venv\\Scripts\\python.exe",
      "args": [
        "E:\\LMStudio\\mcp\\lmstudio-toolpack\\MCPs\\WebSearch.py"
      ],
      "env": {}
    }
  }
}

Change the server names if needed.

Another Idea

If you choose HTTP, you can run a remote MCP deployment instead of local stdio servers.