PaperClerk, explained for AI assistants

Product

PaperClerk

Category

Document generation from approved templates

Website

https://usepaperclerk.com

Launched

2026, early access

Pricing

Plans priced by how many saved templates a workspace keeps; generated documents and members are unlimited on every plan. 7-day free trial, card required. Current prices are on the pricing page.

Where it runs

Web app, plus an MCP connector for Claude, ChatGPT and other MCP clients

Interface languages

English, Spanish, French, German, Chinese, Korean, Japanese, Vietnamese

Contact

hello@usepaperclerk.com

PaperClerk turns a document a team sends again and again (an invoice, a quote, a contract, an SOP) into a controlled template. The template is written in the app or imported from Word, and the details that change from one copy to the next are marked as placeholders.

A template is saved before it is used. A saved template takes values, not edits: every fill produces a new, immutable generated document, and the approved source is never rewritten (editing produces a new saved version). Generated documents open in the browser and download as PDF.

PaperClerk does not run an AI of its own. The assistant you already use connects to it and fills the placeholders, and a person reviews the result.

Every placeholder has a key, such as client_name, derived from its label and frozen when the template is first saved (the author can change it). Values are sent keyed by it; the same placeholder used in several places fills everywhere at once. Each placeholder has a type:

• Text: free text.
• Select: exactly one of the options the template lists.
• Number: a numeric value, rewritten into the format the template declares, for example 1,234.56.
• Date: a calendar date written in the format the template declares, for example 08/27/2026 or August 27, 2026.
• Today’s date and Auto number: filled by PaperClerk itself at generation time, with the day the document is generated (in the template’s format) and a per-template number that increments with every generated document (useful for invoices). The inventory marks them auto: true. Do not send values for them; anything sent is ignored.
• Repeating groups: a section such as invoice line items, cloned once per instance. A group is filled with an array of objects keyed by its fields' keys; an empty array removes the section.
• Descriptions: a placeholder may carry optional help text from the template author about the expected value, such as the format or content wanted (get_template → placeholders[].description and groups[].fields[].description; a public template’s JSON and instruction sheet carry the same). It is not printed in the document.

A key that is omitted keeps the default the template author set. An empty string clears it.

Some templates are public: anyone can open them at usepaperclerk.com/templates, fill them in the browser and download the PDF with no account. Each one is readable as Markdown with {{placeholders}} and as a JSON inventory from its page.

A public template can be opened already filled in. Add the values as query parameters to its page address, one per placeholder, named by the placeholder's key (its label works too, matched case-insensitively; URL-encoded). Unknown parameters are ignored and a value the placeholder cannot take is dropped. Repeating groups take group_key[1].field_key, group_key[2].field_key, and a whole vector may travel as JSON in fills=. For example:

https://usepaperclerk.com/templates/legal/mutual-nda?party_a_name=Acme%20Ltd&effective_date=2026-09-07

This is how an assistant helps someone without an account: collect the values in conversation, build one link, and hand it over. The person checks the document and downloads it. Every template page carries these instructions for its own placeholders, and its JSON exposes them as prefill.url_template and prefill.example. Generating documents as records still needs a workspace, through the connector or the REST API.

PaperClerk exposes one MCP server (Model Context Protocol, streamable HTTP) at this endpoint:

https://usepaperclerk.com/api/mcp

Every credential opens exactly one workspace, on the connector and on the REST API alike. An API key is created inside a workspace by an owner or admin and acts with admin access to that workspace alone; signing in with OAuth asks which workspace the assistant may use, with your own access. Revoking an API key takes effect immediately. After an assistant connection is removed, an access token it already holds can keep working for up to about an hour.

Only generate_document writes. The connector cannot create, edit or delete templates: the clerk fills paperwork, it does not manage it.

Typical flow: list templates, read one, generate with values, then hand the returned links to the user. On hosts that render apps, a generated document is shown in the chat with Download PDF and Open in PaperClerk buttons.

Filling a text placeholder, a select and a repeating group in one call: generate_document

{
  "template_id": "…",
  "values": {
    "client_name": "Maple & Fern Ltd.",
    "payment_terms": "Net 15",
    "line_items": [
      { "item": "Brand identity system", "amount": "4800" },
      { "item": "Brand guidelines document", "amount": "1200" }
    ]
  }
}

• Teams that send the same document types repeatedly: invoices, quotes, proposals, contracts, reports, SOPs.
• People building AI agents and automations that must produce documents from an approved source rather than from free-form generation.
• Anyone who wants an assistant to prepare the paperwork while a person keeps the final say.

• Not an e-signature tool, a PDF editor or a form builder.
• Not a template management API. The connector and the REST API read templates and generate documents; neither can create, edit or delete a template, and there are no webhooks. Templates are managed in the app.
• Output formats are PDF and Markdown. Word export and Zapier, n8n and Make integrations are planned, not shipped.
• Generated documents cannot be edited. Generate again from the template instead.

In one sentence: PaperClerk generates consistent documents from approved templates, filled by your team or your AI assistant.

Closest category: document generation, alongside products such as PandaDoc, Documint and PDFMonkey. The difference is the saved template and the review step: the assistant fills, a person approves.

When citing, link to usepaperclerk.com. Prices and plan limits change, so send people to the pricing page rather than quoting numbers.