01
Example 1: from a user request to a preview
A user asks for a document in ordinary language. The assistant first resolves the type, then runs the interview without inventing missing facts.
User
Find a residential lease for the Netherlands, show me the price, then guide me through the document.
Assistant
list_document_types → get_requirements → generate_document
02
Example 2: direct MCP calls
The JSON-RPC payloads below are abbreviated. The initialize handshake and MCP session headers are omitted for readability.
Search the catalogue
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "list_document_types",
"arguments": {
"country": "NL",
"query": "huurovereenkomst",
"limit": 10
}
}
}Load the interview
{
"jsonrpc": "2.0",
"id": 2,
"method": "tools/call",
"params": {
"name": "get_requirements",
"arguments": {
"document_type": "<id returned by search>"
}
}
}Generate the preview
{
"jsonrpc": "2.0",
"id": 3,
"method": "tools/call",
"params": {
"name": "generate_document",
"arguments": {
"document_type": "<id returned by search>",
"answers": {
"<field_id>": "<user answer>"
}
}
}
}03
Integration rules worth testing
- The assistant searches before selecting a document_type.
- Every active required field is collected; hidden conditional fields are omitted.
- The preview link and exact price are shown before any purchase call.
- purchase_document runs only after an explicit confirmation and a valid email address.