Skip to content
Unstract logo

Unstract

Verified

Extract structured JSON from documents using LLMs via MCP.

MCP ServerCloud & DevOpsLocal (stdio) 6.7k
View on GitHub
Updated 2026-06-15

What is the Unstract MCP server?

Unstract lets users define extraction schemas in natural language through Prompt Studio. It connects to providers including OpenAI, Anthropic, and Ollama, then outputs structured data ready for databases or downstream systems.

The platform can be run locally with Docker or deployed to managed cloud. Its MCP integration allows Claude and similar agents to invoke document processing without custom code.

Install & connect

Set up this server, then add it to your MCP client.

Full setup instructions are in the GitHub repository.

Example prompts

Once connected, try asking your AI client:

Extract key fields from this PDF using the Unstract MCP server
Process the attached invoice image and return structured JSON
Run an ETL job on the documents folder with my defined prompt
List available extraction prompts in my Unstract workspace

Security & permissions

Runs locally via stdio and Docker; requires access to document files and LLM API keys stored in the environment.

What you can do with Unstract

Invoice Processing

Extract line items, totals, and vendor details from invoices into JSON for accounting systems.

KYC Document Analysis

Pull identity and address fields from scanned IDs and forms for compliance workflows.

Insurance Claims Intake

Convert claim forms and supporting documents into structured records for claims platforms.

How to use Unstract

  1. 1Clone the Unstract repository and ensure Docker is installed.
  2. 2Run ./run-platform.sh to start all services.
  3. 3Log in at the local frontend with default credentials.
  4. 4Create a prompt in Prompt Studio and deploy it.
  5. 5Connect your AI client to the MCP server over stdio.

Unstract: pros & cons

Pros

  • +Natural language schema definition speeds up new document types
  • +Supports multiple LLM providers and deployment modes
  • +MCP integration brings extraction into agent workflows
  • +Clean JSON output reduces post-processing effort

Cons

  • Requires Docker and at least 8 GB RAM to run locally
  • Enterprise features need separate licensing
  • No built-in OCR accuracy guarantees for poor scans
Did you find this helpful?

Frequently asked questions

It uses stdio for local communication with AI clients.

User reviews

Verified reviews from the community shape this listing's rating.

Loading reviews…

Sign in to review

Promote Unstract

Add this badge to your website, or share the tool.

DFeatured on DhanasviUnstract 1