Extending public sector intelligence with Agentforce and AWS

Public sector agencies handle vast quantities of unstructured evidence, such as video footage, photographs, and scanned documents, which contain valuable insights that must be extracted before any action can be taken. This article demonstrates how to integrate Amazon Bedrock Data Automation with the Model Context Protocol (MCP) to convert raw data into structured intelligence.

This intelligence can then be accessed via natural language queries within an AI agent like Salesforce Agentforce. In a previous post, Modernizing evidence management in Salesforce Public Sector Solutions with Amazon S3, the integration of Agentforce Public Sector with Amazon S3 was introduced, providing a foundation for storing and managing large files.

However, without automation, significant manual effort is required to classify and extract relevant information from these files. By combining Agentforce with AWS services, users can search for processed data, uncover key insights from unstructured data, and perform advanced actions—all within the Salesforce environment, without the need to leave the console.

The solution consists of two primary workflows. The first workflow involves transferring unstructured media files and documents into Amazon S3 through the External Storage of Files with Amazon S3 for Public Sector connector. This process leverages Amazon S3's scalable storage infrastructure to accommodate large volumes of data. Figure 1 depicts the integration between Agentforce Public Sector and Amazon S3, illustrating how Amazon S3 serves as the backbone for storing extensive documents and media files.

Following the initial storage, an event-driven architecture is employed to process the multimodal data in an asynchronous manner using Amazon Bedrock Data Automation. Figure 2 illustrates the architecture where data stored in Amazon S3 triggers an S3 event notification, which in turn invokes an AWS Lambda function. This function generates a document ID, stores metadata in Amazon DynamoDB, and initiates an Amazon Bedrock Data Automation job to process the file.

The Bedrock Data Automation system extracts structured insights tailored to the file's media type. Upon completion, an Amazon EventBridge rule activates a second Lambda function that saves the results to a designated Amazon S3 output bucket. On the Salesforce side, when a user in Agentforce submits a query, it triggers a configured action that communicates with AWS via MCP.

This communication travels through Amazon Bedrock AgentCore Gateway, a component of Amazon Bedrock AgentCore, which authenticates the request and invokes an MCP server hosted on AWS Lambda. Amazon Bedrock AgentCore is a platform designed for building, connecting, and optimizing AI agents at scale, compatible with any framework or model.

The Lambda function first queries the DynamoDB table to identify the relevant results and then retrieves them from Amazon S3. These results are returned through AgentCore Gateway to Agentforce, where they are loaded into the agent’s context for a natural language response. The architecture is designed to be modular and adaptable, serving as a template that can be tailored to various agency needs beyond evidence management.

It allows for the customization of processing logic within the AWS Lambda MCP Serverless Runtime, the addition of extra metadata in Amazon DynamoDB, and the integration with different agent interfaces via MCP, all without altering the primary data pipeline. To implement this solution, follow the technical implementation guide provided in the linked GitHub repository.

This guide outlines the deployment of the necessary AWS infrastructure, including the use of the AWS Cloud Development Kit (AWS CDK) to set up an event-driven architecture. It also includes instructions for configuring Salesforce Agentforce to connect to the MCP endpoint, contingent upon ensuring that your Salesforce org supports the registration and calling of external MCP servers through the Agentforce Registry. This capability is available across Developer, Enterprise, Performance, and Unlimited Editions.