Dingo MCP Server

What is Dingo MCP Server?

Dingo MCP Server is a Python-based server that exposes Dingo evaluation capabilities through the Model Context Protocol (MCP). It allows developers to run rule-based and LLM-based data quality assessments directly from MCP-compatible clients like Cursor. This server is intended for users who need to evaluate datasets using Dingo’s built-in rule groups or custom LLM configurations.

How to use Dingo MCP Server?

Clone the repository, install dependencies (e.g., pip install -r requirements.txt), then run python mcp_server.py which starts the server via SSE transport by default. Customize the host, port, and log level inside the script’s mcp.run() call. Configure your MCP client (e.g., Cursor) by adding a "url" entry in its mcp.json that matches the server’s address.

Key features of Dingo MCP Server

• Lists available Dingo rule groups and LLM model identifiers.
• Runs rule-based evaluations with configurable rule groups.
• Runs LLM-based evaluations with custom configuration files.
• Supports local, Hugging Face, and other dataset inputs.
• Allows saving detailed outputs (JSONL, correct data) to disk.
• Integrates seamlessly with Cursor’s MCP system.

Use cases of Dingo MCP Server

• Evaluate local JSONL datasets using a predefined set of Dingo rules.
• Perform LLM-based quality checks on text columns with a custom config.
• Automate dataset evaluation directly from an AI coding assistant like Cursor.
• Run batch scoring with configurable concurrency and output paths.

FAQ from Dingo MCP Server

What transport does Dingo MCP Server use by default?

It uses Server-Sent Events (SSE) as the default communication protocol, but this can be changed in mcp.run().

How do I configure the server address for my MCP client?

In your client’s mcp.json, set the "url" to the server’s host and port (e.g., http://127.0.0.1:8888/sse) matching values from mcp.run().

What are the prerequisites to run this server?

You need Git, Python 3.8+, and the fastmcp package. The dingo package must be importable from the cloned repository.

What if my data uses a column name other than 'content'?

Pass the column key via the column_content argument in the kwargs dictionary of run_dingo_evaluation.

How do I provide API keys for LLM evaluations?

API keys must be included inside the custom_config argument (as a file path, JSON string, or dictionary) when calling the LLM evaluation tool.