Dingo
AI data quality evaluation server exposing Dingo's rule- and LLM-based checks over MCP.
Add to your client
Copy the config for your MCP client and paste it into its config file.
pip install dingo-pythonPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"dingo": {
"command": "dingo",
"args": [
"serve",
"--transport",
"stdio"
],
"env": {
"OPENAI_API_KEY": "your-api-key",
"OPENAI_BASE_URL": "https://api.openai.com/v1",
"OPENAI_MODEL": "gpt-4"
}
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- Python with the dingo-python package installed (pip install dingo-python)
- An LLM API key (e.g. OPENAI_API_KEY) for LLM-based evaluations; not needed for rule-based evaluations
- An MCP client such as Cursor or Claude Desktop
About Dingo
Dingo's MCP server exposes its data quality evaluation engine to MCP clients like Cursor and Claude Desktop. It provides six tools to run rule-based or LLM-based evaluations on files, list and inspect Dingo's rule groups / LLM models / prompts, and run quick goal-driven evaluations. Install with pip install dingo-python, then run dingo serve (SSE, default port 8000) or dingo serve --transport stdio. For stdio, the client spawns the process automatically using the command/args config; for SSE, start the server first and point the client at the url (e.g. http://localhost:8000/sse).
Tools & capabilities (6)
run_dingo_evaluationRuns a rule-based or LLM-based Dingo evaluation on an input file or directory. Takes input_path, evaluation_type ('rule' or 'llm'), optional eval_group_name, output_dir, task_name, save_data/save_correct flags, and a kwargs dict (column mappings, data_format, dataset, custom_config with API keys, etc.). Returns the absolute path to the primary output file (e.g. summary.json).
list_dingo_componentsLists available Dingo rule groups, registered LLM model identifiers, and prompt definitions. Accepts component_type ('rule_groups', 'llm_models', 'prompts', or 'all') and include_details. Returns a dict of the requested component lists.
get_rule_detailsGets detailed information about a specific Dingo rule by rule_name, including its description, parameters, and evaluation characteristics.
get_llm_detailsGets detailed information about a specific Dingo LLM by llm_name, including its description, capabilities, and configuration parameters.
get_prompt_detailsGets detailed information about a specific Dingo prompt by prompt_name, including its description, associated metric type, and which groups it belongs to.
run_quick_evaluationRuns a simplified evaluation from a high-level goal. Takes input_path and a natural-language evaluation_goal (e.g. 'check for inappropriate content', 'evaluate text quality'). Returns a summary of results or a path to the detailed results.
When to use it
- Evaluate LLM training / fine-tuning datasets for text quality, completeness, and security directly from an AI IDE
- Run rule-based quality checks (30+ built-in rules) on local JSONL/CSV/TXT/Parquet files via natural-language prompts in Cursor
- Assess RAG system outputs with faithfulness, answer relevancy, and context precision/recall metrics
- Detect hallucinations and run factuality checks on generated content
- Discover and inspect Dingo's available rule groups, LLM evaluators, and prompts before configuring an evaluation
Security notes
LLM-based evaluations require an LLM API key. In stdio mode it is passed via env vars (e.g. OPENAI_API_KEY) in the client config; in SSE mode and for rule customization it is supplied via a custom_config JSON file. Rule-based evaluations need no key. The SSE transport runs an HTTP service (default http://localhost:8000/sse) that any local client can reach — bind host/port carefully if exposing beyond localhost.
Dingo FAQ
Which transport should I use?
SSE is the default (run `dingo serve`, then point the client at http://localhost:8000/sse). stdio is for Claude Desktop or local agent spawn — the client launches `dingo serve --transport stdio` itself, so you don't start the server manually.
Do I need an API key?
Only for LLM-based evaluations. Rule-based evaluations run without any key. For LLM evals, provide the key via env vars (stdio mode) or a custom_config JSON file.
Do I need to clone the repo?
No. `pip install dingo-python` installs the `dingo` CLI which includes the MCP server. Cloning is only needed if you want to customize mcp_server.py directly.
Alternatives to Dingo
Compare all alternatives →
Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.
Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.
Fully managed remote server for AWS docs, blogs, What's-New and Well-Architected guidance — no key.