opc-excel-clean-mcp

AI-Powered Excel/CSV Data Cleaning — 10x Faster Than Kutools Manual Toolbox

Stop clicking through Kutools' $49/year manual toolbox. Let AI clean your spreadsheets in seconds.

Why opc-excel-clean-mcp?

| | opc-excel-clean-mcp | Kutools | 方方格子 | |---|---|---|---| | Speed | ⚡ AI-driven, sub-second | 🐢 Manual 47+ clicks | 🐢 Manual ribbon UI | | Fuzzy Dedup | ✅ Levenshtein distance | ❌ Exact only (manual) | ❌ Limited | | Anomaly Detection | ✅ IQR + Z-score auto-detect | ❌ None | ❌ None | | Batch Processing | ✅ Programmatic (API) | ❌ One sheet at a time | ❌ One sheet at a time | | Cost | 🆓 Free & Open Source | 💰 $49/year (1M+ users) | 🆓 Free (10M+ downloads) | | AI Integration | ✅ MCP native (Claude/GPT) | ❌ None | ❌ None |

Features

1. clean_deduplicate — Remove Duplicates

• Exact dedup: Row-by-row comparison, configurable columns
• Fuzzy dedup: Levenshtein distance catches near-duplicates ("Acme Corp" ≈ "Acme Corporation")
• Control: Choose keep: first or keep: last, set similarity thresholds

2. clean_normalize — Standardize Format

• Trim whitespace, collapse multiple spaces
• Case normalization: lowercase / UPPERCASE / Title Case
• Strip non-printable characters
• Unicode normalization (NFKD)
• Remove empty rows
• Per-column operation targeting

3. detect_anomalies — Find Outliers

• IQR method (robust): Q1 − 1.5×IQR / Q3 + 1.5×IQR
• Z-score method: Mean ± threshold × standard deviation
• Auto-detects numeric columns
• Returns full statistics: Q1, Q3, IQR, mean, median, std dev, bounds

4. fuzzy_match — Fuzzy String Search

• Rank up to 50,000 candidates by Levenshtein similarity
• Configurable threshold and top-N
• Perfect for product name reconciliation, address matching, data merging

Quick Start

Install

npm install -g opc-excel-clean-mcp

Configure in Claude Desktop / Hermes

{
  "mcpServers": {
    "excel-clean": {
      "command": "npx",
      "args": ["opc-excel-clean-mcp"]
    }
  }
}

Use via MCP

User: Clean this sales data — remove duplicates and find outliers

Claude: [calls clean_deduplicate → calls detect_anomalies → returns cleaned data]

5 duplicate rows removed. 3 anomalies found in "Revenue" column:
  - Row 142: $1,250,000 (upper bound: $998,500)
  - Row 287: -$45,000 (lower bound: $12,300)
  - Row 403: $2,100,000 (upper bound: $998,500)

API Reference

All tools accept a data parameter: string[][] — a 2D array where data[0] is the header row and data[1..] are data rows.

clean_deduplicate

| Parameter | Type | Default | Description | |---|---|---|---| | data | string[][] | required | Input tabular data | | mode | "exact" \| "fuzzy" \| "both" | "both" | Dedup strategy | | columns | number[] | all | Column indices to compare | | fuzzy_threshold | number | 0.85 | Similarity (0-1) for fuzzy | | keep | "first" \| "last" | "first" | Which row to keep |

clean_normalize

| Parameter | Type | Default | Description | |---|---|---|---| | data | string[][] | required | Input tabular data | | operations | string[] | ["trim","collapse_whitespace"] | Operations to apply | | columns | number[] | all | Target columns |

Available operations: trim, collapse_whitespace, lowercase, uppercase, title_case, remove_empty_rows, strip_non_printable, normalize_unicode

detect_anomalies

| Parameter | Type | Default | Description | |---|---|---|---| | data | string[][] | required | Input tabular data | | columns | number[] | auto-detect | Numeric columns to scan | | method | "iqr" \| "zscore" | "iqr" | Detection method | | threshold | number | 1.5 | IQR multiplier / Z-score cutoff | | min_values | number | 5 | Min values to analyze |

fuzzy_match

| Parameter | Type | Default | Description | |---|---|---|---| | query | string | required | Search string | | candidates | string[] | required | Pool to search (max 50K) | | top_n | number | 5 | Results to return | | threshold | number | 0.3 | Min similarity (0-1) | | include_scores | boolean | true | Include similarity scores |

Architecture

src/
├── index.ts           # MCP server entry point, tool registration
└── tools/
    ├── fuzzy.ts       # Levenshtein distance, ratio, fuzzy groups, ranking
    ├── dedup.ts       # clean_deduplicate: exact + fuzzy row dedup
    ├── normalize.ts   # clean_normalize: whitespace/case/unicode ops
    └── anomaly.ts     # detect_anomalies: IQR + Z-score outlier detection

• Pure TypeScript — zero runtime dependencies beyond MCP SDK + Zod
• No external API calls — all logic runs locally
• O(n) memory Levenshtein — single-row DP optimization handles large datasets
• Stdio MCP transport — works with any MCP-compatible client

Comparison: AI vs Manual Toolbox

Scenario: Clean 10,000-row sales export with duplicate customers and missing values.

| Step | Kutools (Manual) | opc-excel-clean-mcp (AI) | |---|---|---| | Remove duplicates | Navigate ribbon → Kutools → Select range → Check columns → OK (5 clicks) | clean_deduplicate (1 call) | | Find fuzzy duplicates | Manually scan, sort, eye-ball differences (10+ min) | clean_deduplicate(mode="fuzzy") (<1s) | | Normalize whitespace | Trim + Clean formulas, column by column (20 clicks) | clean_normalize (1 call) | | Detect outliers | Add formulas, conditional formatting, manual review (15 min) | detect_anomalies (<1s) | | Match names | VLOOKUP + manual correction (30+ min) | fuzzy_match (<1s) | | Total | ~45 minutes + 50+ clicks | ~3 seconds + 4 tool calls |

Development

git clone https://github.com/nous-hermes/opc-excel-clean-mcp
cd opc-excel-clean-mcp
npm install
npm run build
npm start

Test locally

echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | node dist/index.js

License

MIT © Nous Research — part of the OPC (One-Person Company) MCP toolkit.