DocumentPro MCP Server

Connect an AI agent to DocumentPro's document extraction and classification platform via the Model Context Protocol — no parsing pipeline to build, no template guessing.

DocumentPro reads invoices, purchase orders, receipts, and tax forms and returns typed JSON keyed to an exact field schema. The MCP server is a thin layer over the same REST API used by DocumentPro's direct API customers — same auth, same metering, same extraction pipeline. Anything an agent does through MCP behaves identically to the REST API.

• Endpoint: https://api.documentpro.ai/mcp

• Transport: Streamable HTTP (stateless, JSON responses)

• Auth: DocumentPro API key in the x-api-key header — create one free

• Billing: tool calls consume credits exactly like the equivalent REST calls, under your existing plan

Connecting

Claude Code

claude mcp add --transport http documentpro https://api.documentpro.ai/mcp \
  --header "x-api-key: YOUR_API_KEY"

Claude Desktop / JSON-config clients

{
  "mcpServers": {
    "documentpro": {
      "type": "http",
      "url": "https://api.documentpro.ai/mcp",
      "headers": { "x-api-key": "YOUR_API_KEY" }
    }
  }
}

Python (MCP SDK)

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async with streamablehttp_client(
    "https://api.documentpro.ai/mcp",
    headers={"x-api-key": "YOUR_API_KEY"},
) as (read, write, _):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await session.list_tools()

Related MCP server: Invoice Parser MCP

Tools

11 tools across four groups.

Extraction

Templates (extraction schemas)

Template schemas look like:

{
  "fields": [
    {"name": "invoice_number", "type": "text", "description": "The invoice number"},
    {"name": "total", "type": "number"},
    {"name": "line_items", "type": "table", "subFields": [
      {"name": "description", "type": "text"},
      {"name": "amount", "type": "number"}
    ]}
  ]
}

Field names: lowercase letters/digits/underscores/spaces (max 50 chars, unique). Types: text, number, date, radio, checkbox, boolean, object, table. table/object require subFields. Every field is nullable. Optional description (max 500 chars) and enum improve accuracy.

Classification

New documents are OCR'd automatically before classification; classify_document waits briefly, and returns DOCUMENT_NOT_READY with a document_id if OCR outlasts the wait — retry with that document_id after ~15 seconds.

Account

Example agent flows

Extract: list_templates → get_schema → extract_document(file_url=...) → poll check_extraction_status(request_id) until completed.

Classify then route: create_classifier(labels=[invoice, purchase_order, other]) once → for each inbound file, classify_document(classifier_id=..., file_url=...) → pick the matching template → extract_document.

Author a template from scratch: get_schema on a similar template for a worked example → create_template(title, schema) → extract_document with the new template_id → inspect results → update_template to refine fields.

Errors

Tools return structured errors: {"error_code": ..., "message": ...}.

Limits

• Inline file_base64 uploads: 7 MB decoded max — host larger files and pass file_url.

• Extraction is asynchronous; classification is synchronous (after OCR).

• Tool calls are metered by your API key's usage plan, same as the REST API.

Learn more

• DocumentPro + MCP — quickstart and setup for any MCP client

• Full API documentation — REST reference, guides

• Sign up — free tier, no credit card

• Contact — custom templates, high-volume use

License

MIT — see LICENSE. This repository documents the DocumentPro MCP server; the server implementation itself is closed-source and hosted by DocumentPro.