Frequently Asked Questions (FAQ) about MCP Server

What is MCP Server?

MCP Server is an open-source inference API server for the RWKV language model, built on the web-rwkv engine. It utilizes Vulkan for GPU acceleration, offering a compact and ready-to-use solution compatible with OpenAI’s ChatGPT API.

What are the key features of MCP Server?

Key features include Vulkan-powered acceleration, RWKV optimization, a lightweight and ready-to-use design, OpenAI compatibility, an open-source license, embeddings API, RAG integration, AI Agent capabilities, BNF sampling, parallel inference, and quantization and LoRA support.

What is RWKV?

RWKV (Receptance Weighted Key Value) is a language model architecture that combines the benefits of Transformers and RNNs, resulting in a powerful and efficient model for various AI tasks.

What is Vulkan and why is it important for MCP Server?

Vulkan is a low-overhead, cross-platform API for accessing GPUs. MCP Server uses Vulkan to enable GPU acceleration without requiring CUDA, allowing it to run on a wider range of GPUs, including AMD cards and integrated graphics.

Do I need an NVIDIA GPU to use MCP Server?

No, MCP Server supports any GPU that supports Vulkan, including AMD cards and integrated graphics.

What is OpenAI compatibility and why is it useful?

MCP Server is compatible with the OpenAI’s ChatGPT API interface, which allows you to seamlessly integrate it with existing AI workflows and leverage familiar tools and techniques.

What is RAG integration in MCP Server?

RAG (Retrieval-Augmented Generation) integration allows your AI applications to access and leverage external knowledge sources, enhancing accuracy and context.

What are AI Agent capabilities in MCP Server?

MCP Server provides the foundation for building sophisticated AI agents, allowing you to automate tasks, create intelligent assistants, and develop innovative AI-powered solutions.

What is BNF sampling?

BNF sampling is a unique feature that allows you to constrain the model’s output to a specific format (e.g., JSON, Markdown) using Backus-Naur Form (BNF) grammars.

How do I get started with MCP Server?

You can download pre-built executables from the Release page, download the RWKV model, configure the Config.toml file (optional), and run the server from the command line.

Can I build MCP Server from source?

Yes, you can build MCP Server from source by installing Rust, cloning the repository, downloading the RWKV model, and compiling the code using Cargo.

Where can I download the RWKV model?

You can download the RWKV model from V5 or V6.

What is UBOS and how does it relate to MCP Server?

UBOS is a full-stack AI Agent development platform that provides the tools and infrastructure needed to orchestrate AI Agents, connect them with data, and customize them with your own LLM models. MCP Server is an asset available on the UBOS marketplace that you can use in your AI Agent development.

How do I convert .pth models to .st format?

You can use the convert_ai00.py or convert_safetensors.py Python scripts (requires Python, torch, and safetensors) or the converter executable available in the Release page.

What APIs are currently available?

The available APIs include /api/oai/v1/models, /api/oai/models, /api/oai/v1/chat/completions, /api/oai/chat/completions, /api/oai/v1/completions, /api/oai/completions, /api/oai/v1/embeddings, and /api/oai/embeddings.