GoldenCheck includes an MCP (Model Context Protocol) server for Claude Desktop and other MCP-compatible clients.

Remote Server (no install required)

GoldenCheck is available as a hosted remote MCP server on Smithery. Connect from Claude Desktop, Claude Code, or any MCP client without installing anything locally.

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "goldencheck": {
      "url": "https://goldencheck-mcp-production.up.railway.app/mcp/"
    }
  }
}

Or browse on Smithery: https://smithery.ai/servers/benzsevern/goldencheck

Local Install

pip install goldencheck[mcp]

Local Setup (Claude Desktop)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "goldencheck": {
      "command": "goldencheck",
      "args": ["mcp-serve"]
    }
  }
}

Restart Claude Desktop. You’ll see GoldenCheck tools available in the tools menu.

Available Tools

scan

Scan a data file for quality issues. Returns findings with severity, confidence, affected rows, and sample values.

Parameters: | Parameter | Type | Required | Default | Description | |———–|——|———-|———|————-| | file_path | string | Yes | — | Path to CSV, Parquet, or Excel file | | llm_boost | boolean | No | false | Enable LLM enhancement | | llm_provider | string | No | “anthropic” | “anthropic” or “openai” | | sample_size | integer | No | 100000 | Max rows to sample |

Returns: Health grade, score, finding count, and full findings list.

validate

Validate a file against pinned rules in goldencheck.yml.

Parameters: | Parameter | Type | Required | Default | Description | |———–|——|———-|———|————-| | file_path | string | Yes | — | Path to data file | | config_path | string | No | “goldencheck.yml” | Path to config |

Returns: Pass/fail status, finding count, and validation findings.

profile

Get column-level statistics: type, null%, unique%, min/max, top values, detected formats.

Parameters: | Parameter | Type | Required | Default | Description | |———–|——|———-|———|————-| | file_path | string | Yes | — | Path to data file | | sample_size | integer | No | 100000 | Max rows to sample |

Returns: Health grade, column profiles with statistics.

health_score

Quick A-F grade for a data file.

Parameters: | Parameter | Type | Required | Description | |———–|——|———-|————-| | file_path | string | Yes | Path to data file |

Returns: Grade (A-F), numeric score (0-100), error/warning counts.

get_column_detail

Deep-dive into a specific column with full statistics and all findings.

Parameters: | Parameter | Type | Required | Description | |———–|——|———-|————-| | file_path | string | Yes | Path to data file | | column | string | Yes | Column name to inspect |

Returns: Full column profile including patterns, enum values, and all findings.

list_checks

List all available profiler checks and what they detect. No parameters.

Example Usage in Claude

“Scan my sales data at /data/sales_2024.csv and tell me what issues you find”

Claude will call the scan tool and present the findings in a readable format.

“What’s the health score of /data/customers.parquet?”

Claude calls health_score for a quick summary.

“Validate /data/orders.csv against my rules”

Claude calls validate to check against goldencheck.yml.

CLI

You can also start the server manually:

goldencheck mcp-serve

This starts the stdio-based MCP server. It’s primarily designed to be launched by MCP clients like Claude Desktop.

Protocol

GoldenCheck supports two transport protocols:

• stdio (local): The server reads JSON-RPC messages from stdin and writes responses to stdout. Used when launched by MCP clients like Claude Desktop via the command config.
• Streamable HTTP (remote): The hosted server at https://goldencheck-mcp-production.up.railway.app/mcp/ uses Streamable HTTP transport. Used when connecting via the url config.

Both follow the MCP specification.

You can also run the HTTP transport locally:

goldencheck mcp-serve --transport http --port 8100