AWS Shows How to Extract Text from PDFs in S3 Using AI Protocol

AWS published a technical how-to guide on its Artificial Intelligence blog titled "Build interactive PDF text extraction from Amazon S3." The post, authored by Phani Parcha and Saibal Gosh and released on 26 June 2026, walks through building a server that extracts text from PDF files stored in Amazon S3 in real time. The tutorial uses the Model Context Protocol (MCP), an open standard that creates a structured communication layer between AI assistants and external data sources. The interactive client demonstrated throughout is Kiro CLI, an AWS-built AI coding tool.

The guide provides a 13-step implementation walkthrough covering four core architecture components: the Kiro CLI client, the MCP protocol layer, a custom MCP server built with Python, and Amazon S3 secured using AWS IAM and CloudTrail.

The approach appeals to businesses that regularly process text-based PDFs stored in cloud storage and want to integrate document processing into their AI workflows. The method works with any text-based PDF but does not handle scanned images, tables, complex form layouts, or documents requiring optical character recognition (OCR). For those use cases, AWS recommends its own Amazon Textract service instead.

The tutorial targets technical teams and developers already using AWS infrastructure who want to reduce extraction costs or build custom document processing pipelines connected to AI assistants. It demonstrates how companies can leverage modern AI protocols to automate document workflows without expensive managed services.

The AWS post provides indicative cost comparisons for processing approximately 10,000 text-based PDF pages per month. Using the traditional Amazon Textract path costs around $23 to $28 monthly, covering Textract charges (approximately $15), S3 storage ($2), Lambda compute ($1), and AI model tokens ($5–$10). By contrast, the MCP server approach costs roughly $2.50 per month, primarily from S3 storage ($2) and data transfer ($0.50).

The article includes a disclaimer that all cost figures are illustrative and may change, and directs readers to official AWS pricing pages for current rates. A typical 50-page text-based PDF generally returns results within a few seconds, with processing time scaling linearly as documents grow larger.

Building this setup requires Python 3.10 or later, Kiro CLI, AWS CLI access, the boto3 library, PyPDF2 for PDF handling, and the mcp Python package. The architecture relies on AWS S3 for document storage and uses AWS IAM for access control and CloudTrail for audit logging.

The guide is categorised as an Advanced (300-level) technical how-to, indicating it targets developers with existing AWS experience rather than beginners. Prerequisites assume familiarity with Python, command-line tools, and AWS service basics.

The tutorial reflects AWS's broader investment in the Model Context Protocol as a production standard. In May 2026, AWS announced general availability of the AWS MCP Server, described as a managed remote protocol server that gives AI agents and coding assistants secure, authenticated access to AWS services. The MCP ecosystem has expanded beyond AWS, with open-source servers for document loading and other data sources now available through AWS Labs.

This PDF extraction guide shows MCP moving from announcement to practical implementation, with step-by-step examples for developers. The post is one of the first real-world walkthroughs of MCP in action on AWS infrastructure, establishing the protocol as a genuine option for production document workflows rather than a theoretical standard.

Stories like this one highlight a growing reality for Australian mid-market businesses: as AI tools become capable of understanding and acting on business documents, having those documents accessible, cleanly organised, and connected to your broader data systems becomes essential. Whether you extract PDFs from S3, invoices from your accounting software, customer records from your CRM, or production schedules from your construction management tool, the value of AI integration depends on whether that data can flow together seamlessly. A unified data foundation that brings PDFs, structured records, and real-time information into one place—audit-ready and permission-controlled—is what turns scattered documents into actionable intelligence that AI can actually use. The lower costs and faster processing times AWS describes are real benefits, but the bigger win is building infrastructure where text extraction, data analysis, and automation work together as one system rather than disconnected point solutions.