Awesome MCP › Software Development

sammcj/mcp-llm

⭐ 78 JavaScript repository created 2025-03-06

MCP LLM is a small MCP server that gives an MCP client access to another language model through the LlamaIndexTS library. The idea is delegation: the assistant a user talks to can hand a sub-task to a different model, for example a cheaper model for boilerplate or a reasoning model for documentation, and receive the result as a tool response. Four tools are documented. The code generation tool takes a natural language description and a target language and returns the generated code. A second tool does the same but writes the output directly into a file, taking a file path, a line number at which to insert, and a count of lines to replace, so generated code can be spliced into an existing source file rather than pasted by hand; the path may be absolute or relative, and relative paths resolve against the server working directory. The documentation tool takes a code snippet, its language and a documentation format such as JSDoc, and returns generated documentation for it. The question tool takes a question and optional context and returns a plain answer, which the README illustrates with a beginner question about JavaScript variable declarations. Installation is either automatic through Smithery, which registers the server with Claude Desktop in one command, or manual from source by cloning the repository, installing dependencies with npm, building, and adding the server to the client MCP configuration. The repository includes an example script that starts the server and drives it with requests, useful for checking the setup outside a chat client. The project is written in JavaScript, released under the MIT licence, and is aimed at developers who want a second model reachable as a tool from inside their existing assistant.

https://github.com/sammcj/mcp-llm

javascriptmcp-serverllamaindexcode-generationdocumentationllm-delegation