LibreChat Code Interpreter MCP Server

A Model Context Protocol (MCP) server that wraps the LibreChat Code Interpreter API, exposing code-execution and file-management endpoints as MCP tools. Easily integrate sandboxed code execution into any MCP-compatible client.

Features

• executeCode — Run arbitrary code (Python, JavaScript, TS, Go, C, C++, Java, PHP, Rust, Fortran, R, D, F90) in a secure sandbox
• downloadFile — Retrieve binary or text files generated during execution
• uploadFiles — Send files (binary or text) to a session for inclusion in code runs
• getFilesInfo — List metadata for files in a session (names, sizes, timestamps)
• deleteFile — Remove a file from a session

All operations require an API key from LibreChat Code Interpreter and adhere to MCP’s tools/list and tools/call schemas

Prerequisites

• Node.js ≥ 16
• npm or yarn
• A valid LibreChat API key (set in LIBRECHAT_API_KEY)

Installation

# Clone this repo
git clone <your-repo-url>
cd <your-repo-folder>

# Install dependencies
yarn install   # or `npm install`

Configuration

Create a .env in the project root containing:

LIBRECHAT_API_KEY=sk-...your_api_key_here...

Running the Server

# If using ts-node:
npx ts-node src/librechat-mcp-server.ts

# Or compile then run:
tsc && node dist/librechat-mcp-server.js

By default, the server communicates over stdio (suitable for local MCP playgrounds, IPC, or Docker streams)

MCP Client Quick Start

Below is an example using the official MCP TypeScript SDK to call executeCode:

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
import { ListToolsResultSchema, CallToolResultSchema } from "@modelcontextprotocol/sdk/types";

async function main() {
  const transport = new StdioClientTransport({ command: "node dist/librechat-mcp-server.js" });
  const client = new Client({ name: "example-client", version: "1.0.0" }, {});
  await client.connect(transport);

  // List available tools
  const tools = await client.request(
    { method: "tools/list" },
    ListToolsResultSchema
  );
  console.log("Available MCP tools:", tools.tools.map(t => t.name));

  // Execute Python code
  const codeResult = await client.request(
    {
      method: "tools/call",
      params: {
        name: "executeCode",
        arguments: {
          code: "print(\"Hello from LibreChat!\")",
          lang: "py",
          user_id: "user123",
          entity_id: "assistant_ABC123",
          files: [],
        },
      },
    },
    CallToolResultSchema
  );

  console.log("Execution result:", codeResult.content[0].json);
}

main().catch(console.error);

Tool Schemas

| Tool Name | Input Parameters | | ------------ | --------------------------------------------------------------------------- | | executeCode | code (string), lang (enum), args?, user_id?, entity_id?, files? | | downloadFile | session_id (string), fileId (string) | | uploadFiles | entity_id (string), files (array of { name, content, contentType }) | | getFilesInfo | session_id (string), detail? (simple or detailed) | | deleteFile | session_id (string), fileId (string) |

Combine these via MCP’s standard tools/list and tools/call messages to build powerful code-automation flows

Development

• Lint: npm run lint
• Type-check: npm run build
• Format: npm run format

Feel free to open issues or pull requests!