Prem

You run a Model Context Protocol (MCP) server to expose Prem AI capabilities—like chat completions, retrieval-augmented generation, document management, templates, and streaming responses—so MCP clients can interact with Prem AI in a consistent, scalable way.

How to use

With an MCP client, connect to Prem MCP Server to perform chat operations, retrieve and incorporate documents for RAG, upload documents to repositories, and apply predefined templates for specialized outputs. Start the server in stdio mode to enable real-time streaming and robust error handling as you query models, manage documents, and control output through templates.

How to install

Prerequisites: Node.js (v16 or higher), a Prem AI account with API key, and a Prem project ID.

npx -y @smithery/cli install @ucalyptus/prem-mcp-server --client claude

Additional sections

Configuration and usage details are provided below to help you set up and operate Prem MCP Server in your environment, including environment variables, example runtime configurations for MCP clients, usage scenarios, and troubleshooting steps.

The server supports uploading documents to Prem AI repositories for RAG operations and accepts common document formats such as .txt, .pdf, and .docx for repository indexing and retrieval.

Starting and using the server in stdio mode enables you to run the MCP server as a local process and connect MCP clients through standard input/output streams. This mode is especially suitable for desktop clients like Claude Desktop and Cursor. Ensure your API key and project ID are provided as environment variables when launching the server.

If you plan to manage multiple repositories or use templates, you can reference template IDs and repository IDs to tailor responses, control response length, and specify similarity thresholds for document matching.

Notes on configuration and usage

Environment variables control access to Prem AI and scope of operations. You will typically provide your API key and project ID in a dedicated environment file or inline in the runtime configuration for MCP clients.

When using Cursor, configure the MCP server with a stdio entry that points to the server entry point and passes credentials through environment variables.

For Claude Desktop, set up a similar stdio-based MCP connection, including the required environment variables for authentication and project scoping.

Example prompts and usage patterns

Basic Chat: Initiate a conversation with the Prem AI model to explore capabilities and responses.

RAG with documents: Ask questions that leverage repository content and document chunks for more informed answers.

Using templates: Apply predefined prompt templates to produce outputs in a specific style or format.

Document upload

You can upload documents to Prem AI repositories for RAG operations. Supported formats include .txt, .pdf, and .docx.

Available tools

Chat Completions

Interact with Prem AI language models to generate responses, follow conversation context, and handle user prompts.

RAG Support

Retrieve and incorporate information from document repositories to augment model responses.

Document Management

Upload, organize, and manage documents within repositories for later retrieval.

Template System

Apply predefined prompt templates to produce outputs in specific styles or formats.

Streaming Responses

Receive model outputs in real time as they are generated.

Error Handling

Provide robust error handling and logging to improve reliability and observability.