AGENT INFRASTRUCTURE
Build agents that write, sign, and validate documents
AI agents can read and understand documents. They can’t render PDFs, apply digital signatures, validate compliance, or fill form fields. Nutrient provides the document infrastructure layer — deterministic, auditable, production-ready — so your agents can do real work with real documents.
Autodesk
IBM
UBS
Dropbox
Disney
Lufthansa
Autodesk
IBM
UBS
Dropbox
Disney
Lufthansa
THE READ-WRITE GAP
AI agents can read documents. They can’t write them.
Large language models excel at understanding, summarizing, and extracting information from documents. But they can’t render a PDF page, cryptographically sign a contract, validate PDF/A compliance, fill interactive form fields, or perform pixel-accurate redaction. These operations require purpose-built document infrastructure — deterministic, auditable, and battle-tested. Nutrient bridges this gap.
WHAT LLMs CAN DO
Read and understand
Summarize and extract text
Parse content, answer questions, and pull structured data from documents.
Classify and categorize
Identify document types, detect intent, and route to the right workflow.
Translate and compare
Convert between languages and surface semantic differences between versions.
WHAT LLMs CAN’T DO
Write and manipulate
Render, sign, and validate
Pixel-perfect PDF rendering, PAdES digital signatures, PDF/A compliance validation.
Fill forms and flatten
Populate interactive form fields, flatten annotations, apply watermarks.
Redact with precision
Permanent, irrecoverable PII removal — not black boxes layered on top of text.
Nutrient MCP: Natural-language document operations
Hand your LLM a prompt. Get a finished, compliant document back. No SDKs to learn, no API calls to write — just infrastructure that works.
Redact all PII from these three contracts, sign them, and convert to PDF/A-3.
Extract the invoice table. Then merge all receipts into a single report.
OCR this scanned form and extract the field values as JSON.
GET STARTED IN MINUTES
Connect your agent to Nutrient
- Add Nutrient MCP to Claude Code or Codex with the command shown here. Nothing to install.
- Sign in with your Nutrient account in the browser. No API key needed.
-
In the Claude app, add
https://mcp.nutrient.io/mcpas a custom connector. In other MCP clients, add it as a remote MCP server.
# Claude Codeclaude mcp add --transport http --scope user nutrient https://mcp.nutrient.io/mcp# Then run /mcp, choose nutrient, and select Authenticate
# Codexcodex mcp add nutrient --url https://mcp.nutrient.io/mcp && codex mcp login nutrientWORKS WITH EVERY AGENT FRAMEWORK
Claude, GPT, LangGraph, and more
Claude Code/Anthropic tool use
Add the hosted server to Claude Code with one command and start prompting. The same MCP tool layer also fits Anthropic-powered workflows in your own runtime.
OpenAI Agents/GPT function calling
Expose the server’s JSON schema as function tools. Deterministic responses, no hallucinated API calls.
LangGraph/LangChain
Wire Nutrient operations as tool nodes in your graph. Each operation returns structured JSON.
Custom agent frameworks
Connect to the hosted server over HTTP from any framework with MCP client support.
What your agents can do
Document operations your agents can run: deterministic, auditable, and production-ready.
AVAILABLE NOW
Hybrid document intelligence in the Data Extraction API
Document understanding that combines vision language models with traditional algorithmic OCR. Table extraction, key-value pairing, and intelligent character recognition, delivered by a hybrid pipeline with the accuracy of proven algorithms and the flexibility of modern AI. Purpose-built for the documents AI agents struggle with most: scanned forms, handwritten notes, and complex multicolumn layouts. Shipping today in the Data Extraction API’s understand and agentic modes, with per-field confidence and source grounding.
AI document assistant
Embed an AI assistant directly inside your document viewer. Users ask questions, summarize, translate, redact, and extract — in context, on the document they’re looking at.
- In-document Q&A with source references
- AI-powered redaction with human in the loop
- Summarize, translate, and classify on selection
Web, iOS, and Android SDKs
Embed AI Assistant into any app on any platform with a few lines of code.
Bring your own LLM
OpenAI, Anthropic, AWS Bedrock, or self-hosted open source models — your choice.
Semantic document comparison
AI-enhanced diffs that surface meaning-level changes, not just character differences.
TRUSTED AT SCALE
Infrastructure that enterprises depend on
Nutrient processes millions of documents daily for organizations that can’t afford to get document handling wrong.
15%
of the Global 500 trust Nutrient for document infrastructure
130+
public sector organizations across 24 countries
80+
nations where Nutrient-powered documents are used
SOC 2
Type 2 audited with GDPR and CSA compliance
SECURITY AND DEPLOYMENT
Your infrastructure, your rules
Document compliance is infrastructure, not intelligence. Deploy Nutrient on your terms — self-hosted, managed cloud, or hybrid — with the security controls your organization requires.
You control what’s kept
Nutrient MCP keeps nothing itself. Extraction runs are saved to your run history until you delete them, and uploaded files expire after 24 hours.
Self-hosted or managed cloud
Deploy on your infrastructure with your LLM, or use Nutrient’s managed cloud with dedicated SLAs.
Your data stays yours
Nutrient never trains models on your documents. Self-hosted LLMs keep everything within your perimeter.
Audit-ready by default
Every operation is logged and traceable. SOC 2 Type 2 audited, GDPR-compliant, and built for regulated industries.
REGULATED INDUSTRIES
Need isolated, compliance-ready document agents?
For enterprises in healthcare, legal, finance, and government — dedicated document processing agents with full audit trails, HIPAA/SOC 2/GDPR compliance, and zero access to anything but documents.
Give your AI agents document superpowers
Try the SDKs and Document Engine free, or connect your agent to Nutrient MCP with a free account.
Frequently asked questions
What is Nutrient MCP and how does it work with AI agents?
Nutrient MCP is a hosted Model Context Protocol (MCP) server at https://mcp.nutrient.io/mcp. Your agent says “redact PII and sign this PDF,” and the server runs the document operations on Nutrient’s document APIs. Add the URL to Claude, Claude Code, Codex, or another MCP client that supports remote servers, and sign in with your Nutrient account.
What is the read-write gap and why does it matter?
The read-write gap is the fundamental limitation of AI agents when working with documents. LLMs excel at reading — understanding, summarizing, and extracting information from documents. But they can’t write — render PDFs, apply cryptographic signatures, validate compliance standards, fill interactive forms, or perform pixel-accurate redaction. These write operations require deterministic, purpose-built document infrastructure. Nutrient bridges this gap by providing the infrastructure layer that makes AI agents capable of complete document workflows.
Which agent frameworks are supported?
Nutrient MCP works with any client or framework that can connect to a remote MCP server:
- Claude — Add the hosted server as a custom connector, sign in, and start prompting.
- Claude Code and Codex — Add the hosted server with one command, sign in, and start prompting.
- OpenAI Agents and GPT — Expose as function tools via JSON schema.
- LangGraph and LangChain — Wire as tool nodes in your agent graph.
- Custom frameworks — Connect to the hosted server over HTTP with an MCP client, and sign in with OAuth.
What document operations can my agent perform?
Through Nutrient MCP, your agent has access to these operations:
- Extraction — Fields described in plain words, returned with citations and confidence.
- Parsing — PDFs, scans, and images to Markdown and tables.
- Classification — Sort documents by type.
- Conversion and OCR — Office to PDF, OCR, merge, rotate, watermark, PDF/A, and PDF/UA.
- Digital signatures — Sign PDFs.
- AI redaction — Find and remove personal data.
- Credits — Check how many credits are left.
The SDKs and Document Engine add more, such as form filling and interactive editing in your own app.
What is Nutrient Data Extraction?
Data Extraction is the hybrid engine that combines vision language models (VLMs) with traditional algorithmic OCR to deliver document intelligence for the documents AI agents struggle with most: scanned forms, handwritten notes, and complex layouts. Its capabilities ship today in the Python and Java SDKs for self-hosted use, and through the Data Extraction API’s understand and agentic modes, including table extraction, key-value pairing, and intelligent character recognition (ICR), with per-field confidence scores. A standalone hosted endpoint is in development; contact Sales to learn more.
How does Nutrient handle security for agent-driven document processing?
Security is built into the infrastructure:
- You control what’s kept — Nutrient MCP keeps nothing itself. Extraction runs are saved to your run history until you delete them, and uploaded files expire after 24 hours.
- Self-hosted option — Deploy on your infrastructure with your own LLM for complete data sovereignty.
- SOC 2 Type 2 — Audited security controls for enterprise deployments.
- No training on your data — Nutrient never uses your documents to train models.
Is Nutrient MCP open source?
No. Nutrient MCP isn’t open source. The earlier open source local MCP server is no longer supported.
How do I get started?
Two steps:
- Add the server to your agent. In the Claude app, add
https://mcp.nutrient.io/mcpas a custom connector. In Claude Code, runclaude mcp add --transport http --scope user nutrient https://mcp.nutrient.io/mcp. In Codex, runcodex mcp add nutrient --url https://mcp.nutrient.io/mcp && codex mcp login nutrient. - Sign in with your Nutrient account in the browser when your agent asks.
For setup steps for each client, see the Nutrient MCP page or developer documentation.