Aparavi MCP Server

A Python Model Context Protocol (MCP) server that integrates with the Aparavi Platform. It exposes Aparavi data processing tools to LLMs, allowing them to stream local files directly to Aparavi EaaS for advanced processing (OCR, classification, PII detection, etc.).

Features

• Full DTC capabilities accessible within your MCP client
• Custom pipeline tools created from your DTC workflows
• Built-in tools for immediate use (no DTC configuration required)
• Document processing and text extraction
• OCR capabilities for images and system diagrams
• Async processing support
• MCP-compliant interface

Installation

Option 1: Using pipx (Recommended)

Isolate the server environment:

pipx install aparavi-mcp

Option 2: Using pip

Install into your current environment:

pip install aparavi-mcp

Configuration

Add the server to your MCP host configuration (e.g., mcp.json for Cursor, claude_desktop_config.json for Claude Desktop).

You will need your Aparavi API key and URI.

{
  "mcpServers": {
    "aparavi": {
      "command": "aparavi-mcp",
      "args": [],
      "env": {
        "APARAVI_URI": "https://eaas.aparavi.com",
        "APARAVI_AUTH": "your-api-key-here"
      }
    }
  }
}

Note: For EU users, please use https://eaas.aparavi.eu for APARAVI_URI.

Getting Your API Key

To obtain your API key for authentication, visit your Aparavi account settings or contact your Aparavi administrator.

Usage

Once configured, the following tools will be available to your LLM:

1. Aparavi Document Processor

A convenience tool that processes a file using a standard Aparavi document processing pipeline.

• Input: filepath (Absolute path to the local file)
• Output: Extracted text and metadata

2. Custom Pipelines

Any running tasks/pipelines in your Aparavi account will also be dynamically exposed as tools.

Pipeline Requirements

To use custom pipelines as tools, they must follow these configuration rules in DTC:

Webhook Start

Your pipeline must start with a webhook source node.

HTTP Response End (for client output)

If you need output returned to your MCP client, your pipeline must end with an HTTP response node.

HTTP Response Exception

Pipeline output to your client may not always be necessary. For example, a pipeline that sends data to a vector store node won't require an HTTP response node, as it stores data rather than returns it to the client.

Example Pipeline Flow

For client output:

Webhook → [Your Processing Nodes] → HTTP Response

For vector storage:

Webhook → [Your Processing Nodes] → Vector Store

Once your pipeline is configured in DTC with a webhook source, make sure to start the pipeline. It will automatically become available as a tool in your MCP client.