Document AI MCP Server
Natural‑language PDF operations — prepared for LLMs
Hand your LLM a prompt and get a finished, compliant PDF back. MCP Server plugs Claude, GPT, or any agent framework straight into Nutrient’s battle‑tested DWS APIs, so document workflows run on words, not glue code. One‑line install, Docker‑friendly, and streams over HTTP or stdio.
.png)
Watch the live demo
Watch this demo of what Nutrient MCP Server is capable of.
Built for the LLM era
SDK sidekicks — GitHub Copilot snippets and the upcoming llms.txt metadata surface the right Nutrient calls directly in your IDE.
Agent‑ready interfaces — Deterministic JSON schema keeps LangGraph, OpenAI Agents, and Claude Desktop predictable.
Sandboxed execution — Every job runs in an isolated workspace; secrets and file systems stay safe.
Stateless streaming — Long‑running PDF operations stream back in real time, so your agents never hit token limits. We prioritize meaningful results over busyness.
Business outcome — Launch AI features months faster while slashing integration risk.
Examples of what you can ask
Zero‑code — no scripts or custom code required.
“Convert my DOCX to PDF.”
DOCX → PDF conversion via DWS; file returned to sandbox
“Convert all PDFs to DOCX.”
Batch-converts all PDFs in the directory
“Extract the table from invoice.pdf and print it.”
Table extraction model returns structured JSON
“Perform key-value extraction on my credit card application.”
Auto-detects a file and returns names, dates, and SSNs
“Sign invoice.pdf and watermark it with PAID.”
Adds a visible watermark, embeds a PAdES signature
Core capabilities
Digital signatures and watermarks
PAdES‑compliant signatures and branded overlays in one shot.
Implementation in three simple steps
Install and launch locally:
npx nutrient-mcp-server --sandbox ./sandbox --port 4000
Ask Claude something like:
Convert all PDFs to DOCX and save them.
Watch files appear in ./sandbox
Prefer Docker?
docker run -p 4000:4000 ghcr.io/nutrient/mcp-server:latest.
API performance you can trust
Nutrient MCP Server works with Nutrient DWS Processor API — fast, predictable document processing at scale.
Documents processed a day
Average latency
From first call to “done”
requests per minute (test key rate) and 5 MB file size limit
Coming soon
Document Engine MCP inside the container
Natural‑language control baked into every on‑premises deployment, zero extra microservices.
Native Windows build, and one‑click PowerShell install
Put zero‑code PDF workflows on internal desktops and RPA bots.
Drag‑and‑drop agent templates
Prewired chains for contract review, onboarding, and compliance you can remix in minutes.
Frequently asked questions
What is Nutrient MCP Server?
Nutrient MCP Server is an open source microservice that translates natural language prompts from any large language model (LLM) into deterministic JSON calls to Nutrient Document Web Services (DWS) APIs. It then streams back finished PDFs, structured JSON, or images — no glue code required.
How do I call MCP Server from GPT‑4o, Claude, or LangGraph agents?
You expose the server’s JSON schema as a function/tool and let the LLM choose the right operation. Examples for OpenAI “function calling,” Anthropic “tool” messages, and LangGraph nodes live in the repository.
Do I really need zero code?
Yes. Spin up a sandbox with npx nutrient-mcp-server or the Docker image, and then prompt: “Convert all PDFs to DOCX and save them.” The agent does the rest — no scripts, SDKs, or configuration files.
How does MCP Server keep documents and secrets secure?
Every job runs in an isolated sandbox directory; streams are stateless and can travel over TLS. No files or tokens ever leave the workspace unless you explicitly export them.
Which document operations ship out of the box?
Conversion, merge, split, OCR, AI redaction (PII/PHI/PCI), table and key‑value extraction, PDF/A‑UA, PAdES signatures, watermarks, and more all work out of the box, matching Nutrient’s full DWS feature set.
Is MCP Server really open source?
Yes — refer to the MIT license on GitHub. Fork it, extend it, or file an issue.
Where can I run it and what are the requirements?
You can run it on Linux (x86-64) or Apple Silicon with Node 20 or newer. It’s also easy to deploy with a single command on Docker or Kubernetes. If you’re on Windows, a native build and PowerShell installer are on the way and currently in final testing.
What’s coming next and why should I care?
Soon, the Document Engine container will ship with MCP baked in, giving every on‑premises customer natural language control by default. Drag‑and‑drop agent templates for contract review, onboarding, and compliance will follow — slashing your time‑to‑value even further.