Goldenflow MCP Server

Moved. This repo has moved into the benzsevern/goldenmatch monorepo at packages/python/goldenflow (and packages/typescript/goldenflow)/. This repo is archived; new development happens in the monorepo.

GoldenFlow

Data transformation toolkit — standardize, reshape, and normalize messy data before it hits your pipeline. Built by Ben Severn.

Works on files (CSV, Excel, Parquet), cloud storage (S3, GCS), or live databases. Zero-config mode auto-detects what needs fixing. One command to clean what GoldenCheck found and prep what GoldenMatch needs. Available in Python and TypeScript with full feature parity.

# Python
pip install goldenflow
goldenflow transform data.csv

# TypeScript / Node.js
npm install goldenflow
npx goldenflow-js transform data.csv

---

The Problem

Your data arrives broken in predictable ways:

• Phone numbers come in 15 different formats
• Dates are mixed between MM/DD/YYYY and YYYY-MM-DD
• Addresses have inconsistent abbreviations
• Column names don't match between systems ("fname" vs "first_name" vs "given_name")
• Values have leading whitespace, unicode garbage, smart quotes
• Categoricals are inconsistent ("USA", "US", "United States")

Every data engineer writes throwaway scripts to fix these. Every script is slightly different. None of them are reusable.

GoldenFlow makes the transforms reusable, composable, and automatic.

---

Quick Start

Python

pip install goldenflow

# Auto-transform (zero-config)
goldenflow transform messy_data.csv

# Try the demo first
goldenflow demo
goldenflow transform demo_data.csv -c demo_config.yaml

# With config
goldenflow learn messy_data.csv -o config.yaml
goldenflow transform messy_data.csv -c config.yaml

# Schema mapping
goldenflow map --source system_a.csv --target system_b.csv

# Full pipeline
goldencheck scan data.csv
goldenflow transform data.csv
goldenmatch dedupe data_transformed.csv

TypeScript / Node.js

npm install goldenflow

# Auto-transform (zero-config)
npx goldenflow-js transform messy_data.csv

# Try the demo first
npx goldenflow-js demo
npx goldenflow-js transform demo_data.csv -c demo_config.yaml

# With config
npx goldenflow-js learn messy_data.csv -o config.yaml
npx goldenflow-js transform messy_data.csv -c config.yaml

# Schema mapping
npx goldenflow-js map -s system_a.csv -t system_b.csv

Programmatic (TypeScript)

import { TransformEngine } from "goldenflow";

const result = new TransformEngine().transformDf([
  { name: "  JOHN  ", phone: "(555) 123-4567", email: "John@Example.COM" },
]);
console.log(result.rows[0]);
// { name: "JOHN", phone: "+15551234567", email: "john@example.com" }

---

Zero-Config Mode

goldenflow transform customers.csv
# or just:
goldenflow customers.csv

GoldenFlow profiles every column and applies safe transforms automatically:

• Strips whitespace and normalizes unicode
• Standardizes phone numbers to E.164 format
• Normalizes email casing
• Parses and standardizes date formats to ISO 8601
• Normalizes zip codes (zero-padding, strip +4)
• Replaces smart/curly quotes with straight quotes
• Auto-corrects categorical misspellings via fuzzy matching

Output: a clean CSV with a sidecar manifest showing every transform applied.

customers.csv           -> customers_transformed.csv
                        -> customers_manifest.json

The manifest is an audit trail — what changed, why, and which rows were affected.

---

CLI Commands

GoldenFlow has 14 commands. The most common ones:

# Core transforms
goldenflow transform data.csv                    # Zero-config: auto-detect and fix
goldenflow transform data.csv -c config.yaml     # Apply saved config
goldenflow transform data.csv --domain healthcare # Use a domain pack
goldenflow transform data.csv --strict           # Fail on any transform error
goldenflow transform data.csv --llm              # Enable LLM-enhanced corrections
goldenflow data.csv                              # Shorthand: auto-routes to transform

# Schema & profiling
goldenflow map -s a.csv -t b.csv                 # Auto-map schemas between files
goldenflow profile data.csv                      # Show column profiles
goldenflow learn data.csv -o config.yaml         # Generate config from data patterns
goldenflow validate data.csv                     # Dry-run: show what would change
goldenflow diff before.csv after.csv             # Compare pre/post transform

# Continuous & scheduled
goldenflow watch ./data/                         # Auto-transform new/changed files
goldenflow schedule data.csv --every 1h          # Run on a schedule (5m, 1h, 30s...)
goldenflow stream large_file.csv --chunk-size 50000  # Stream-process in batches

# Discovery & history
goldenflow init data.csv                         # Interactive setup wizard
goldenflow demo                                  # Generate sample data to try
goldenflow history                               # Show recent transform runs
goldenflow history -n 50                         # Last 50 runs

# Integrations
goldenflow interactive data.csv                  # Launch TUI
goldenflow serve                                 # REST API for real-time transforms
goldenflow mcp-serve                             # MCP server for Claude Desktop

Default Routing

Running goldenflow <file> without a subcommand auto-routes to transform:

goldenflow customers.csv       # equivalent to: goldenflow transform customers.csv
goldenflow -                   # read from stdin, write to stdout

---

Streaming

For files too large to load into memory, use StreamProcessor or the stream command:

goldenflow stream large_file.csv --chunk-size 50000

from goldenflow.streaming import StreamProcessor

processor = StreamProcessor(config=config)

# Process a single record
result = processor.transform_one({"name": "  John  ", "phone": "(555) 123-4567"})

# Process a batch
result = processor.transform_batch(df_batch)

# Stream a large file in chunks
for result in processor.stream_file("large_data.csv", chunk_size=10_000):
    write_to_output(result.df)

print(f"Processed {processor.batches_processed} batches")

---

Cloud Connectors

GoldenFlow reads from and writes to S3 and Google Cloud Storage transparently:

# S3
goldenflow transform s3://my-bucket/raw/customers.csv -o s3://my-bucket/clean/

# GCS
goldenflow transform gs://my-bucket/data/records.csv

from goldenflow.connectors.s3 import read_s3, write_s3
from goldenflow.connectors.gcs import read_gcs, write_gcs

df = read_s3("s3://my-bucket/raw/customers.csv")
df = read_gcs("gs://my-bucket/data/records.csv")

Cloud paths are detected automatically — no extra flags needed.

---

Watch Mode

Auto-transform files as they arrive in a directory:

goldenflow watch ./data/
goldenflow watch ./incoming/ -c config.yaml -o ./processed/

GoldenFlow polls the directory and applies transforms to any new or changed files.

---

Scheduling

Run transforms on a repeating schedule:

goldenflow schedule data.csv --every 1h
goldenflow schedule data.csv --every 30m -c config.yaml -o ./output/

Supported intervals: 30s, 5m, 1h, 2h, etc.

---

Setup Wizard

Generate a YAML config interactively:

goldenflow init data.csv

The wizard profiles your data, suggests transforms, and saves a goldenflow.yaml ready to use.

---

History

GoldenFlow tracks every transform run in ~/.goldenflow/history/:

goldenflow history         # Last 20 runs
goldenflow history -n 50   # Last 50 runs

Each run record captures: source file, row count, transforms applied, errors, and duration.

---

Schema Mapping

When you need to merge data from different systems:

goldenflow map --source crm_export.csv --target warehouse_schema.csv

GoldenFlow auto-maps columns between schemas using name similarity and data profiling:

crm_export.csv              warehouse schema
-----                       -----
email_address      ->       email (rename)
phone_number       ->       phone (rename)
fname              ->       first_name (alias match)
st                 ->       state (alias match)

Ambiguous mappings get flagged for human review. Confident mappings apply automatically.

---

Domain Packs

Pre-configured transform sets for common industries. All 5 are now implemented:

goldenflow transform patients.csv --domain healthcare
goldenflow transform employees.csv --domain people_hr
goldenflow transform transactions.csv --domain finance
goldenflow transform orders.csv --domain ecommerce
goldenflow transform listings.csv --domain real_estate

---

Transform Library (76 transforms)

Text Transforms (18)

Phone Transforms (5)

Name Transforms (8)

Address Transforms (8)

Date Transforms (13)

Categorical Transforms (6)

Numeric Transforms (9)

Email Transforms (4)

Identifier Transforms (3)

URL Transforms (2)

---

Special Modes

Strict Mode

Fail immediately if any transform error occurs — useful in CI or production pipelines:

goldenflow transform data.csv --strict

Exits with code 1 and prints the first 5 errors if any transform fails.

LLM Mode

Use an LLM to enhance categorical corrections and handle edge cases that fuzzy matching misses:

goldenflow transform data.csv --llm

Requires OPENAI_API_KEY or ANTHROPIC_API_KEY in your environment. Falls back to standard transforms gracefully.

Auto-Correct

category_auto_correct uses fuzzy matching to fix misspelled categorical values automatically. It is suppressed on high-cardinality columns (>10% unique values) to avoid false positives.

"actve" -> "active"
"Pennsylvnia" -> "Pennsylvania"
"Unted States" -> "United States"

---

YAML Config

For repeatable pipelines:

# goldenflow.yaml
source: customers.csv
output: customers_clean.csv

transforms:
  - column: name
    ops: [strip, title_case]
  - column: email
    ops: [lowercase, strip]
  - column: phone
    ops: [phone_e164]
  - column: state
    ops: [state_abbreviate]
  - column: signup_date
    ops: [date_iso8601]

renames:
  email_address: email
  phone_number: phone

drop: [internal_id, temp_notes]

dedup:
  columns: [email]
  keep: first

goldenflow transform customers.csv -c goldenflow.yaml

Generate a config from your data automatically:

goldenflow learn data.csv -o config.yaml

---

Python API

import goldenflow

# Zero-config
result = goldenflow.transform_file("messy_data.csv")
print(result.df)          # Clean Polars DataFrame
print(result.manifest)    # Audit trail

# With config
from goldenflow import GoldenFlowConfig, TransformSpec, TransformEngine

config = GoldenFlowConfig(
    transforms=[
        TransformSpec(column="phone", ops=["phone_e164"]),
        TransformSpec(column="date", ops=["date_iso8601"]),
    ]
)
engine = TransformEngine(config=config)
result = engine.transform_df(df)

Jupyter Notebook Support

TransformResult, Manifest, and DatasetProfile all have _repr_html_() — they render as rich HTML tables automatically in Jupyter:

import goldenflow

result = goldenflow.transform_file("messy_data.csv")
result           # renders as HTML table in Jupyter
result.manifest  # renders transform audit trail

---

TypeScript / JavaScript

GoldenFlow has a full TypeScript port with feature parity — same 83 transforms, same engine, same config format. The core is edge-safe (runs in browsers, Cloudflare Workers, Vercel Edge) with a Node layer for file I/O and CLI.

Install

npm install goldenflow

CLI

npx goldenflow-js transform data.csv              # Zero-config
npx goldenflow-js transform data.csv -c config.yaml  # With config
npx goldenflow-js profile data.csv                 # Column profiles
npx goldenflow-js learn data.csv -o config.yaml    # Generate config
npx goldenflow-js diff before.csv after.csv        # Compare files
npx goldenflow-js map -s source.csv -t target.csv  # Schema mapping
npx goldenflow-js stream large.csv --chunk-size 50000  # Streaming
npx goldenflow-js demo                             # Generate sample data
npx goldenflow-js history                          # Show recent runs

TypeScript API

import { TransformEngine, makeConfig } from "goldenflow";

// Zero-config — auto-detect and fix
const engine = new TransformEngine();
const result = engine.transformDf([
  { name: "  John Smith  ", email: "JOHN@EXAMPLE.COM", phone: "(555) 123-4567" },
  { name: "DR. JANE DOE", email: "  jane+work@gmail.com  ", phone: "555.987.6543" },
]);

console.log(result.rows);
// [
//   { name: "John Smith", email: "john@example.com", phone: "+15551234567" },
//   { name: "Jane Doe", email: "jane@gmail.com", phone: "+15559876543" },
// ]
console.log(result.manifest.records.length); // transforms applied

// Configured — explicit transforms per column
const engine = new TransformEngine({
  transforms: [
    { column: "phone", ops: ["phone_e164"] },
    { column: "email", ops: ["strip", "email_normalize"] },
    { column: "name", ops: ["strip", "title_case"] },
    { column: "state", ops: ["state_abbreviate"] },
    { column: "price", ops: ["currency_strip"] },
    { column: "signup_date", ops: ["date_iso8601"] },
  ],
  renames: { email_address: "email" },
  drop: ["internal_id"],
  dedup: { columns: ["email"], keep: "first" },
});
const result = engine.transformDf(rows);

// Schema mapping
import { SchemaMapper } from "goldenflow";

const mapper = new SchemaMapper();
const mappings = mapper.map(
  [{ fname: "John", lname: "Smith", email_address: "j@e.com" }],
  [{ first_name: "", last_name: "", email: "" }],
);
// [{ source: "fname", target: "first_name", confidence: 0.95 }, ...]

// Streaming large datasets
import { StreamProcessor } from "goldenflow";

const processor = new StreamProcessor({ transforms: [{ column: "name", ops: ["strip"] }] });
for (const result of processor.streamRows(largeDataset, 10_000)) {
  await writeChunk(result.rows);
}

// Profiling
import { profileDataframe } from "goldenflow";

const profile = profileDataframe(rows, "customers.csv");
for (const col of profile.columns) {
  console.log(`${col.name}: ${col.inferredType}, ${col.nullCount} nulls, ${col.uniqueCount} unique`);
}

// Edge-safe import (browsers, Workers, Edge Runtime)
import { TransformEngine } from "goldenflow/core";

// Node-only import (includes file I/O, MCP, CLI)
import { readFile, TransformEngine } from "goldenflow/node";

MCP Server (TypeScript)

import { TOOL_DEFINITIONS, handleTool } from "goldenflow/node";

// TOOL_DEFINITIONS: 10 MCP tools for Claude Desktop
// handleTool("transform", { path: "data.csv" }) → JSON string

REST API (TypeScript)

import { runServer } from "goldenflow/node";
runServer(8000); // Starts HTTP server with /health, /transforms, /transform

---

Public API (34 exports)

from goldenflow import (
    # Core engine
    TransformEngine, TransformResult,
    # Config
    GoldenFlowConfig, TransformSpec, SplitSpec, FilterSpec, DedupSpec, MappingSpec,
    # Convenience
    transform_file, transform_df,
    # Manifest
    Manifest, TransformRecord, TransformError,
    # Profiler
    DatasetProfile, ColumnProfile,
    # Selector & differ
    select_transforms, diff_dataframes, DiffResult,
    # Transform registry
    TransformInfo, register_transform, get_transform, list_transforms, parse_transform_name,
    # Mapping
    SchemaMapper, ColumnMapping,
    # Config helpers
    load_config, save_config, merge_configs, learn_config,
    # Domains
    DomainPack, load_domain,
    # Connectors
    read_file, write_file,
)

---

Integrations

REST API

goldenflow serve --host 0.0.0.0 --port 8000

POST CSV data, get transformed CSV back. Built with FastAPI.

MCP Server

goldenflow mcp-serve

Exposes GoldenFlow as an MCP tool for Claude Desktop. Configure in your Claude Desktop settings.

TUI

goldenflow interactive data.csv

Full-featured terminal UI built with Textual. Browse profiles, apply transforms, preview results.

Remote MCP Server

GoldenFlow is available as a hosted MCP server on Smithery — connect from any MCP client without installing anything.

Claude Desktop / Claude Code:

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

Local server:

pip install goldenflow[mcp]
goldenflow mcp-serve

10 tools available: transform files, auto-map schemas, profile columns, generate configs, diff before/after, apply domain packs.

---

Part of the Golden Suite

Raw Data
   |
   v
+--------------+
|  GoldenCheck |  <- Discover quality issues
|  goldencheck |
|  scan data   |
+------+-------+
       | findings
       v
+--------------+
|  GoldenFlow  |  <- Fix issues, standardize, reshape
|  goldenflow  |
|  transform   |
+------+-------+
       | clean data
       v
+--------------+
|  GoldenMatch |  <- Deduplicate, match, create golden records
|  goldenmatch |
|  dedupe      |
+------+-------+
       | golden records
       v
   Clean, deduplicated,
   production-ready data

Chain them:

goldencheck scan data.csv | goldenflow transform --from-findings | goldenmatch dedupe

---

Performance

Built on Polars (Rust-backed DataFrames). Transforms use a hybrid approach: native Polars expressions stay in the Rust engine for simple transforms (strip, lowercase), while complex transforms (phone parsing, date parsing) use optimized Python via map_batches.

---

Benchmarks

GoldenFlow scores 100/100 on the DQBench transform benchmark across all three tiers (customer database, e-commerce, healthcare claims).

pip install dqbench
dqbench run goldenflow

---

Why GoldenFlow?

GoldenFlow is purpose-built for the transform step between validation and matching — not a general ETL tool. It turns messy data into clean, standardized data automatically.

---

Error Handling

GoldenFlow catches errors at the CLI boundary and shows friendly, actionable messages — no raw stack traces. Individual transform errors are captured in the manifest rather than crashing the run. Use --strict to change this behavior.

---

Progress Bars

Long-running operations (streaming, watch mode, scheduling) display a Rich progress spinner showing batch count, rows processed, and estimated completion.

---

GitHub: github.com/benzsevern/goldenflow Author: Ben Severn License: MIT Python: 3.11+ | Node.js: 20+ | npm: goldenflow | PyPI: goldenflow