JustFill MCP Server

Let your AI agent fill PDF forms.

JustFill speaks Model Context Protocol. Claude, ChatGPT, n8n — any MCP client can detect form fields on any PDF or scan, verify them visually, fill the document and save a template so the next fill is instant.

No install: add it as a connector

JustFill runs a hosted MCP server with OAuth — in Claude (Settings → Connectors → Add custom connector) or ChatGPT (developer mode), just point at:

https://justfill.app/api/mcp

Your agent gets a browser consent screen on first use — no keys to copy, revocable anytime in Account → API Keys. Works with Gemini CLI too: gemini extensions install https://github.com/mrmaciej1/justfill-mcp

n8n: add an MCP Client Tool node to your AI Agent, set the endpoint to https://justfill.app/api/mcp and authenticate with a Bearer header using an API key from Account → API Keys. Prefer a no-LLM pipeline? Import the ready-made workflow — PDF + data in, filled PDF out, deterministic via saved templates.

Local install (CLI agents)

  1. 1. Add the server to your MCP client config (Claude Desktop, Claude Code, or any other). The package is on PyPI, so uvx fetches it automatically:

    {
  "mcpServers": {
    "justfill": {
      "command": "uvx",
      "args": ["justfill-mcp"]
    }
  }
}

  2. 2. Authorize once — opens your browser, one click while logged in to JustFill:

    uvx justfill-mcp login

    Prefer explicit configuration? Create a key under Account → API Keys and set JUSTFILL_API_KEY instead. No passwords in config files either way.

  3. 3. Ask your agent — “Fill ~/Downloads/intake-form.pdf with the client data from this email and show me a preview before saving.”

Built for agents that need to be right

Works on scans, not just AcroForms

ML detection finds fields on flat PDFs and phone photos of paper forms. Embedded form fields and saved templates are used first — they are deterministic.

Visual review loop

The agent sees rendered previews with labeled boxes and honest confidence scores, fixes mistakes with field edits, and previews the filled result before a single credit is spent.

Templates make repeat fills exact

Review a form once, save it as a template — every future fill of that document skips detection entirely and lands on verified geometry.

International text done right

Output supports Latin (with full diacritics), Cyrillic, Greek, Arabic and Hebrew with proper shaping and direction, plus Chinese, Japanese, Korean and Vietnamese.

What the agent can do

open_pdf

Open a PDF or a phone-photo scan (JPG/PNG). Fields come from a saved template, embedded AcroForm, or ML detection with calibrated confidence.

render_preview

See the detected boxes drawn on the page — blue is deterministic, green/orange/red is ML confidence.

render_filled_preview

Preview your values in place before filling. Costs nothing.

add / update / remove_field

Fix detection mistakes — including batch versions and prune_fields for bulk cleanup.

fill_pdf

Fill and save the document. Native AcroForm fill when available, precise text overlay otherwise. Warns about values that will not fit.

save_template

Persist the reviewed layout — every future fill of that form is instant and exact.

Open source client, same account

The MCP client is MIT-licensed. It talks to the same JustFill account and credits as the web app — detection is free, fills follow your plan.

PyPI: justfill-mcpGitHub: mrmaciej1/justfill-mcp