How to turn notes or an email into a finished document with an AI agent Leave feedback

Paste an order, an email or meeting notes, and your agent turns them into a finished document: it extracts the facts into the template’s data skeleton, and the GroupDocs.Assembly MCP server — running locally — renders the contract. If the agent forgets a field, the server says which one:

dnx GroupDocs.Assembly.Mcp --yes

Here is the order from Contoso Pharma’s email. Make the service agreement SA-2026-118 from contract-template.docx and save it as SA-2026-118-contoso.docx.

Note

The commands and config snippets on this page are for the .NET build of the server — the only platform available today. Installation and client setup: MCP server for .NET. Other platforms will expose the same tools with their own launch command; everything else on this page applies unchanged.

• The server registered in your AI client — quick start.
• A template whose fields match what the documents usually say — see generate documents with AI agents.

user              agent                        Assembly MCP server
 |  "Make the contract for this order: Contoso Pharma, 12 months ..."  |
 |---------------->|                                                   |
 |                 |-- get_template_fields ------------------------->|
 |                 |   {"template":{"filePath":                        |
 |                 |     "contract-template.docx"}}                    |
 |                 |<- {"dataSourceNames":["contract","customer",      |
 |                 |    "items"], "dataSkeleton":{"customer":{"Name":  |
 |                 |    ""...}, "items":[{"Title":"","Quantity":0}]}} -|
 |                 |   the agent fills the skeleton from the order     |
 |                 |-- assemble_document --------------------------->|
 |                 |   {"template":{...},"dataJson":"{...}",           |
 |                 |    "format":"pdf"}                                |
 |                 |<- "Assembled 'contract-template.docx' with inline |
 |                 |    JSON (JSON object) as PDF" --------------------|
 |<- "Contract SA-2026-118 is ready: 3 services, total 75500 EUR."     |
   the model writes the data; layout, tables and wording come from the template

The email gave the services, the customer and the contact, but not the contact’s e-mail address — so the agent’s first data had no customer.Email. Captured from server 26.11.0 (the evaluation note that opens each response is trimmed):

-> get_template_fields {"template":{"filePath":"contract-template.docx"}}
<- { "fields": [ "contract.Number", "contract.Date", "contract.Discount",
     "customer.Name", "customer.Address", "customer.Contact",
     "customer.Email", "items[].Title", "items[].Quantity",
     "items[].UnitPrice" ], "dataSkeleton": { ... } }

-> assemble_document {"template":{"filePath":"contract-template.docx"},
     "dataJson":"<order without customer.Email>"}
<- Document assembly failed for 'contract-template.docx': ... Can not get the
   value of member 'Email' on type ''. ... The template reads 'Email', which
   the data does not have. Check the spelling (names are case-sensitive) with
   get_template_fields and add it to the data. A list that is empty in the
   data has no fields at all, so reading one fails too; pass
   allowMissingMembers=true to output nothing for missing members and empty
   lists.

-> assemble_document {"template":{"filePath":"contract-template.docx"},
     "dataJson":"<order with customer.Email>",
     "outputName":"SA-2026-118-contoso.docx"}
<- Assembled 'contract-template.docx' with inline JSON (JSON object) as DOCX
   Saved to: …/SA-2026-118-contoso.docx

The right move after the failure is to ask you for the address, not to invent one — say so in your agent instructions (below). allowMissingMembers: true is for fields that may legitimately be empty, not for facts the agent could not find.

SA-2026-118-contoso.docx with the customer block, three service rows, the total of 75500 EUR and the discount line — and, before it, a JSON object you can read in ten seconds to check what the agent extracted.

Is the extraction reliable? It is the model’s reading of your text; that is why the data is a small JSON object you can review before the document exists. The layout and the arithmetic are not the model’s — the template does those.

Can it read the email file itself? Your agent reads it with its own tools (or another GroupDocs server, such as Parser). This server takes the data the agent produced.

Does the email leave my machine? The server reads nothing but the template and writes the document locally. What you paste into the conversation goes to your model provider, as always.

Agents follow project instructions about when to use a tool. Paste this into the instruction file your agent reads (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md, .cursorrules):

When asked to turn notes or an email into a document: call get_template_fields,
fill the dataSkeleton only with facts present in the source, show me the JSON,
and ask for anything missing instead of using allowMissingMembers or guessing.

• Evaluation mode: every generated document starts with an Evaluation Only line, workbooks gain a worksheet, and e-mail templates lose their HTML body. Licensing.
• <<[name]>>, not {{name}}: other engines’ placeholders stay in the output as text.
• Templates are trusted input: image and document tags load the paths and URLs their expressions return.
• Dates: write them as "2026-11-12T00:00:00" and format them in the template; a date-only string can shift by a day on machines east of UTC.

• Generate documents with AI agents
• One document per record (mail merge)
• Reports with tables and charts
• Inspect and fix a template
• Reports from repository data
• On-premise deployment
• Tools reference · Licensing · Troubleshooting & FAQ
• Questions: Assembly forum

• Let the model fill in the data, not write the document — the same scenario on the GroupDocs blog, with worked examples and the trade-offs.