An OCR Model Context Protocol server that hands your AI agent typed JSON, not raw text. Bank statements, invoices, receipts, checks, tax forms, all callable as native tools.
Add this snippet to your MCP client config and restart. Works in Claude Desktop, Cursor, Continue, and Cline.
{
"mcpServers": {
"docuclipper": {
"command": "npx",
"args": ["-y", "docuclipper-mcp"],
"env": {
"DOCUCLIPPER_PAT": "dcp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
}
}
}
}User: I'm dropping in a folder of vendor invoices from Q1. Extract them, group by vendor, and tell me which vendors I spent the most with.
Claude (using docuclipper.convert_document on each):
- Extracted 47 invoices across 12 vendors
- 3 had reconciliation flags (line items + tax did not match printed total)
- Top 3 by spend: Acme Supplies $24,180; CloudCo $18,440; Office Depot $9,712
Want me to flag the 3 reconciliation issues?The same extraction engine as the DocuClipper REST API, exposed as MCP tools.
9,000+ banks. Reconciled transactions, account info, opening/closing balances. Per-bank templates not required.
Vendor, dates, line items, totals, tax. Header fields with 99% accuracy, line items at 95-98%, plus a reconciled flag for low-confidence files.
MICR routing/account, payee, courtesy + legal amount, date, memo for checks. W-2, 1099, K-1, and other tax forms field-level extracted.
Tools are discovered live by Claude Desktop, Cursor, Continue, and Cline. Your agent calls them like any other built-in tool, no orchestration code required.
Not raw OCR text dumps. Agents get typed JSON: dates as ISO strings, amounts as numbers, line items as arrays, balances as decimals. Easy to reason about.
Add command: npx, args: [-y, docuclipper-mcp] to your MCP config. The latest tool registry is fetched on demand.