OpenViking

OpenViking MCP Server enables an AI assistant to access the OpenViking knowledge base through semantic search, document reading, pattern matching, summaries, and full-text search. It exposes a local, configurable server that your MCP client can interact with to retrieve and summarize knowledge assets efficiently.

How to use

You run the OpenViking MCP Server locally and connect to it from your MCP client. The server is designed to perform semantic search over your knowledge base, read documents like PDFs and Markdown files, match files using glob patterns, generate concise summaries, and perform keyword full-text searches. Use the MCP client to invoke the available endpoints, such as semantic search, document reading, glob matching, and summarization, by issuing requests that Target the local MCP server.

How to install

Prerequisites you need before installation include Python and a working MCP client. You will also run supporting services for model embeddings and the local MCP server.

pip install openviking mcp

Configuration and run details

Configure the local MCP server to run as a subprocess of your environment, and point it at your data and configuration file. The server is designed to be started via a Python command and will use environment variables for data path and configuration.

{
  "mcp": {
    "openviking": {
      "type": "local",
      "command": ["python", "path\\to\\openviking_mcp_server.py"],
      "environment": {
        "OPENVIKING_DATA_PATH": "path\\to\\viking_data",
        "OPENVIKING_CONFIG_FILE": "path\\to\\ov.conf"
      },
      "enabled": true,
      "timeout": 20000
    }
  }
}

Usage in your MCP client

From your MCP client, connect to the local server using the provided command and environment. Ensure the process has access to the designated data directory and configuration file. Once connected, you can perform semantic searches, read documents, apply glob patterns to locate files, obtain L0 summaries, L1 overviews, and conduct full-text searches across your knowledge base.

Project structure and available tools

The server project includes modules for semantic search, document reading, glob-based file matching, and summarization. You can review usage examples and internal interfaces to understand how to call each function from your MCP client.

Notes and troubleshooting

Ensure the data path and configuration file are accessible by the server process. If the server fails to start, verify the timeout setting and confirm that Python can locate the script at the specified path. If embeddings or auxiliary services are required, ensure they are running and reachable by the MCP server.

Available tools

openviking_find

Semantic search across the knowledge base using embeddings to locate relevant documents and passages.

openviking_read

Read and retrieve content from documents, supports PDF and Markdown formats.

openviking_glob

Pattern-based file matching using glob patterns to filter datasets.

openviking_abstract

Generate L0-level summaries of documents or results.

openviking_overview

Create higher-level L1 overviews that provide a concise outline of results.

openviking_grep

Full-text keyword search across indexed contents.