MCP Integration
4.7/5(91+ reviews)Trusted by 10,000+ finance teams
Connect DocuClipper to Your AI Agent (MCP)
Plug DocuClipper into Claude Desktop, Cursor, Continue, and other MCP clients. Your agent extracts bank statements and invoices end to end, no manual upload step.
4.7/5(91+ reviews)Trusted by 10,000+ finance teams14-day free trialNo credit card required
Last updated:
4+
MCP clients supported (Claude Desktop, Cursor, Continue, Cline)
5
Tools exposed (one-shot + async building blocks)
99.6%
Extraction accuracy
4.7/5
G2 rating
Built for AI Agents That Need to Read Financial Documents
The docuclipper-mcp package gives any Model Context Protocol client first-class access to DocuClipper extraction, with structured output your agent can actually reason about.
Works with Claude Desktop, Cursor & Continue
The docuclipper-mcp package is a stdio Model Context Protocol server, so any MCP-aware client (Claude Desktop, Cursor, Continue, Cline, and others) can call DocuClipper as a native tool.
One-shot extraction from a chat prompt
Drop a bank statement PDF into Claude Desktop and ask for a CSV. The convert_bank_statement tool uploads, extracts, polls, and returns transactions in a single call.
Bank statements and invoices, structured
Agents get clean per-transaction rows, account-level grouping, and invoice line items, the same structured output our REST API returns. No HTML scraping or screenshot parsing.
Personal Access Token auth
Generate a PAT under Settings > API & Webhooks, drop it in DOCUCLIPPER_PAT, and your agent inherits your DocuClipper credit balance and document history.
Async path for large files
Files over ~10 MB use a presigned-S3 upload flow (upload_url, convert_document, get_job_status, download_transactions). The shim exposes both one-shot and async tools.
No install, just npx
All MCP clients use command: npx, args: [-y, docuclipper-mcp]. No global install, no Python venv, no Docker. Tool registry is discovered live so you always get the latest set.
How It Works
Three steps to wire DocuClipper into Claude Desktop, Cursor, or any MCP-compatible client.
1
Generate a Personal Access Token
In DocuClipper, go to Settings > API & Webhooks and click Create Personal Access Token. Copy the dcp_... token (it is shown only once).
2
Add docuclipper-mcp to your client config
Paste the JSON snippet below into your MCP client's config file (claude_desktop_config.json, .cursor/mcp.json, etc.). Set DOCUCLIPPER_PAT to your token.
3
Restart and prompt
Restart your client. DocuClipper tools appear in the tool menu. Ask your agent to extract a PDF and it calls convert_bank_statement automatically.
Drop-in MCP Client Config
The same shape works in Claude Desktop, Cursor, Continue, and Cline. File location varies by client.
claude_desktop_config.json / .cursor/mcp.jsonJSON
{
"mcpServers": {
"docuclipper": {
"command": "npx",
"args": ["-y", "docuclipper-mcp"],
"env": {
"DOCUCLIPPER_PAT": "dcp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
}
}
}
}Claude Desktop config path
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Cursor config path
Project: .cursor/mcp.json
Global: ~/.cursor/mcp.json
Full setup guide: Connect DocuClipper to AI agents (MCP).
Supported MCP Clients
docuclipper-mcp is a standard stdio MCP server, so any client that speaks the protocol works.
Claude Desktop
Anthropic's desktop client (macOS and Windows). Edit claude_desktop_config.json, restart, and DocuClipper appears under the hammer-icon tools menu.
Cursor
AI-native code editor. Add to .cursor/mcp.json in your project (or globally at ~/.cursor/mcp.json) so your agent can extract documents while you code.
Continue
Open-source IDE assistant for VS Code and JetBrains. Use the same npx-based config in Continue's MCP settings.
Cline & other MCP clients
Any client that speaks Model Context Protocol over stdio works, including Cline, custom agent runtimes, and self-hosted orchestrators.
What Your Agent Can Do
Common workflows finance teams and developers run through the DocuClipper MCP server.
Pull a year of statements into a CSV
Drag 12 monthly PDFs into Claude Desktop and say give me one combined CSV. The agent calls convert_bank_statement on each, merges the rows, and hands back a clean export.
Reconcile invoices in your IDE
From Cursor, ask the agent to extract structured fields from a folder of supplier invoices and write them to a spreadsheet, or to your AP system via your own code.
Build agentic workflows on top of extraction
Chain DocuClipper into a larger agent loop, classify transactions, flag duplicates, draft journal entries. The agent treats DocuClipper as one tool among many.
Reuse already-completed jobs
download_transactions pulls structured data from any historical job by ID, so your agent can answer follow-up questions without re-uploading the same PDF.
What Customers Say
Real reviews from accountants, bookkeepers, and finance teams.
“DocuClipper has helped us eliminate several manual data entry processes, saving us a lot of time.”
KR
Kristin Mitchell
Accounting, United States
“It's a complete game-changer. Instead of spending hours combing through statements, we get the data we need almost instantly.”
MA
Matt
Lending, United Kingdom
“DocuClipper allowed us to enhance our advisory services, directly impacting our bottom line.”
SA
Sarah Winship
Accounting, United Kingdom
Wire DocuClipper into your AI agent in five minutes. Start your free 14-day trial.
Start free trialFrequently Asked Questions about the DocuClipper MCP Server
It is a small Node package, docuclipper-mcp, that implements the Model Context Protocol over stdio. MCP clients like Claude Desktop, Cursor, and Continue spawn it as a subprocess and call DocuClipper tools (convert_bank_statement, download_transactions, upload_url, convert_document, get_job_status) as if they were built into the assistant.
There is nothing to install globally. All MCP clients use command: npx with args: [-y, docuclipper-mcp], so the package is fetched and run on demand. You only need Node.js 18 or later available on the machine running your MCP client.
Sign in to DocuClipper, go to Settings > API & Webhooks (/account?section=api), and click Create Personal Access Token. The token is shown once at creation, copy it into your MCP client config under env.DOCUCLIPPER_PAT.
Bank statements (the default tool surface) and invoices via the standard agent endpoints. Tax forms, receipts, and generic OCR continue to use the legacy /protected/* API and are not currently exposed through the MCP shim.
convert_bank_statement is intended for files under ~10 MB and times out at 300 seconds by default. For large or slow jobs, use the async path: upload_url to get a presigned S3 URL, PUT the file, call convert_document, then poll get_job_status and finish with download_transactions.
The MCP server is a thin wrapper over the same /api/v1/agent/mcp/* HTTP endpoints. If you are building a custom agent runtime that cannot spawn a subprocess, call those endpoints directly. If you are wiring up Claude Desktop, Cursor, or Continue, the stdio shim is the easier path.
No. The web Claude.ai client does not currently support local MCP stdio servers. Use Claude Desktop, Cursor, or another MCP-aware client. For server-side agents, call the HTTP endpoints directly.