KoboldCPP

Kobold MCP Server provides an MCP-compatible interface to KoboldAI, enabling text generation, chat completion with memory, and OpenAI-compatible endpoints. Once running, clients can send requests through the MCP protocol to perform standard text generation, chat with persistent context, or leverage additional capabilities like stable diffusion, audio transcription, and TTS via the KoboldAI integration. The server is implemented in TypeScript and packages the KoboldAI interaction logic behind MCP endpoints, so applications that expect MCP can interact without direct KoboldAI-specific calls. To get started, install the npm package, configure the KoboldAI API endpoint, and launch the server. You can then connect your MCP client to the server-name and issue requests as you would with other MCP servers, including text completion, chat, and memory-enabled conversations. The server also exposes OpenAI-like text completion endpoints to ease integration with tools expecting OpenAI-compatible requests.

Prerequisites:

• Node.js v16 or higher
• npm or yarn
• Access to a KoboldAI instance (running and reachable via HTTP API)

Install the MCP server package:

npm install kobold-mcp-server

If you plan to run from source, clone and build:

git clone https://github.com/yourusername/kobold-mcp-server.git
cd kobold-mcp-server
npm install
npm run build

Configure and run:

1. Create a configuration file or set environment variables to point to your KoboldAI API endpoint.
2. Start the server (the package exports the Kobold MCP server class and you can run via your own Node script or use a runner that imports the package).

Example run with a simple Node script:

// run-server.js
import { KoboldMCPServer } from 'kobold-mcp-server';

const server = new KoboldMCPServer({ apiUrl: 'http://localhost:5001' });
server.start();

Then execute:

node run-server.js

Alternatively, if you expose a CLI in your environment, you can start it using that mechanism after building.

Environment variables and configuration options:

• KOBOLD_API_URL: URL of your KoboldAI API endpoint (required for the server to route requests)
• KOBOLD_API_KEY: Optional API key if your KoboldAI deployment requires authentication
• NODE_ENV: Set to production for optimized runtime Common issues:
• Ensure KoboldAI is running and accessible from the MCP server host
• Verify CORS/settings on the KoboldAI API if you encounter request failures
• If using Stable Diffusion or audio features, confirm the respective services are reachable and properly configured Configuration tips:
• Use a dedicated network or container for KoboldAI and the MCP server to reduce latency
• Monitor logs for request routing errors to identify misconfigured API URLs