Openfilter-MCP

This is a Model Context Protocol (MCP) server that allows instruction-optimized AI agents such as large language models (LLMs) or vision language model (VLM) to access code and information about Plainsight OpenFilter.

By leveraging semantic retrieval capabilities, any off-the-shelf LLM or VLM can efficiently access relevant information and code snippets from the OpenFilter platform, answering questions or planning to ensure that pipelines and filters are optimized for user requirements.

This project uses uv. first, install dependencies;

uv sync

Preindex:

uv run index

And finally, serve:

uv run serve

Docker

You can build and run this with Docker:

# Run the container
docker run --name openfilter-mcp -d -p 3000:3000 plainsightai/openfilter-mcp

# Check logs
docker logs openfilter-mcp

# Stop the container
docker stop openfilter-mcp

The server will be available at:

http://localhost:3000/mcp

The configuring MCP client should then add the MCP server under the HTTP transport to the required url, e.g.:

{
  "mcpServers": {
    "openfilter": {
      "type": "http",
      "url": "http://localhost:3000/mcp"
    }
  }
}

or, for clients without explicit HTTP support:

{
  "mcpServers": {
    "openfilter": {
      "command": "npx", // or equivalent, e.g., `pnpm dlx`
      "args": ["-y", "mcp-remote", "http://localhost:3000/mcp", "--allow-http"]
    }
  }
}