nimonkaranurag/code-buddy

Plug and play code assistant

• MCP(Model Context Protocol) allows a customer to augment their MCP-compliant LLM-applications with additional tools which a server can expose.

• This repo hosts a re-purposed OSS MCP-server which can be integrated into applications like:

  • Claude Desktop
  • Cursor
  • Watson Orchestrate
  • or any application with an MCP client!

• Still in development, future iterations will include the ability to plug-and-play this server; allowing for an enhanced UX!

Problem Statement

Enterprise repos require limited visibility, 
while developers might not be able to share source code 
directly with external services like ChatGPT, Claude, etc. 
because of company policy - often, 
the document sources are public-facing information.

Onboarding a new developer into an established codebase is in itself non-trivial 
and depending on experience, 
this process can be significantly streamlined for 
early-mid career professionals if 
they have access to using 
the aforementioned powerful external services 
for guidance and assistance.

Integration Steps

• This will be streamlined for easy integration into common MCP compliant LLM applications.

• This assistant server will be containerized and exposed as a service to allow for a lightweight runtime integration.

• For now, these resources will prove handy:

  • llms.txt
  • how to connect to local MCP servers

Sample config for Claude Desktop

{
  "mcpServers": {
    "code-buddy": <--- set in `server.py` {
      "command": "/Users/nimo/.local/bin/uv", <-- do `which uv`
      "args": [
        "run",
        "--active",
        "--project", "/Users/nimo/Desktop/wxo/wxo-agent-evaluation/src/wxo_agentic_evaluation/utils/code-buddy", <-- set to absolute path of project root
        "/Users/nimo/Desktop/wxo/wxo-agent-evaluation/src/wxo_agentic_evaluation/utils/eval-assistant/main.py" <-- must be absolute path to the `main.py` of this repo
      ],
      "env": {
        "IBM_GITHUB_ACCESS_TOKEN": "REDACTED"
      }
    }
  }
}

• Use the mcpinspector to investigate potential tool calling errors and for testing.