SV Excel Agent

Last reviewed: 2026-04-28

Autonomous Agent MCP Open Source

Best for: Multi-step Excel tasks that require reasoning — "analyze this data and build a summary sheet" — not just single cell edits but end-to-end autonomous spreadsheet work

Not ideal for: Simple read/write operations — if you just need to read a cell or write a value, the Excel MCP Server (3.6k stars) is more established and widely tested

Autonomous Agent Loop

How the SV Excel Agent operates

OBSERVE Read workbook structure, sheet names, data ranges, and formulas to understand context

↓

THINK Claude reasons about the task: what transformations are needed, which cells to touch, what formulas to write

↓

ACT Write values, apply formulas, create sheets, format ranges — via MCP tool calls

↓

VERIFY Re-read changed cells to confirm edits applied correctly; loop back if adjustments needed

The observe-think-act-verify loop means Claude can complete complex multi-step tasks without human intervention between each step.

Key Capabilities

🧠

Autonomous Multi-Step Execution

Give a high-level task description and the agent plans and executes the steps — no need to decompose into individual read/write operations manually

📊

Full XLSX Manipulation

Read data, write values, create sheets, set formulas, apply formatting, sort/filter data, create charts — full spreadsheet manipulation via MCP

🔗

Works in Claude Desktop

MCP protocol enables use in Claude Desktop — ask Claude to work on your spreadsheet in natural language from the desktop chat interface

✅

Self-Verification

After making changes, the agent reads back the affected cells to verify correctness — reduces silent errors from formula mismatches or wrong cell references

Signals

GitHub Stars

178

Type

Autonomous Agent (MCP)

Works In

Claude Desktop + Code

License

Open Source

Requires

OpenAI API Key

Quality Assessment

Task Autonomy

4.3

Formula Accuracy

4.0

Ease of Use

3.5

Community Testing

2.8

Community testing score is lower due to the smaller star count (178) compared to Excel MCP Server (3.6k) — fewer users have battle-tested edge cases. For production use, validate on your specific workbook structure first.

Install & Use

# Clone and install
git clone https://github.com/SylvianAI/sv-excel-agent
cd sv-excel-agent
pip install -r requirements.txt

# Add to Claude Desktop config
{
  "mcpServers": {
    "excel-agent": {
      "command": "python",
      "args": ["/path/to/sv-excel-agent/server.py"]
    }
  }
}

# Then ask Claude in Claude Desktop:
# "Read sales.xlsx and create a pivot summary by region on a new sheet"
# "Calculate YoY growth in column D using the values in columns B and C"
# "Format the header row bold, freeze the top row, and add conditional formatting"
        