ACST's Secure AI Gateway: A Blueprint for Trust in Data-Driven Ministry

FLORENCE, SC – July 14, 2026 – In a move that directly addresses one of the biggest hurdles to AI adoption in sensitive sectors, church technology provider ACST has announced the general availability of MP-MCP, a secure AI gateway for its MinistryPlatform customers. The new connector allows churches to use powerful AI assistants like Claude to analyze their live data, promising a future where insights are derived from conversations rather than complex reports. "The future of ministry isn't more reports—it's conversations with your data," stated Joe Koehling, CEO of ACST. While the promise of conversational data is compelling, the true innovation here is not the connection itself, but the robust, multi-layered security framework designed to protect highly sensitive congregant information.

A 'Guest, Not Admin' Security Model

For any organization handling personal data, the prospect of connecting a third-party AI to a core database is fraught with risk. For churches, where data includes everything from family relationships and pastoral care notes to donation histories, the standard for data stewardship is even higher. ACST is tackling this challenge with a security philosophy it calls "guest, not admin."

"Churches are eager to embrace AI, but they shouldn't have to compromise security or governance to do it," Koehling explained. The goal is to ensure the AI remains a helpful but restricted guest, never an administrator with full access. This philosophy is enforced through six distinct layers of server-side protection:

1. Permission Parity: The AI inherits the exact permissions of the human user making the query. It can see nothing the user isn't already authorized to see.
2. Granular Control: Administrators can enable or disable specific AI tools based on a user's security role.
3. Default Secrecy: Sensitive data tables are inaccessible to the AI by default, requiring explicit administrative action to be exposed.
4. PII Masking: Personally Identifiable Information (PII) can be automatically masked before any data is sent to the AI model, protecting individual privacy.
5. Read-Only by Default: Write access is disabled, preventing the AI from making unauthorized changes to the database.
6. Full Auditing: All AI activity is logged for complete transparency and oversight.

Crucially, Koehling emphasized that these protections are built into the server-side code. "There is no ‘jailbreak’ prompt that can recover data the server was never willing to send," he said, directly addressing a common vulnerability in AI systems. Furthermore, ACST confirms that congregation data is retrieved live for each query and is never stored by the AI provider or used to train its models—a critical safeguard for maintaining data privacy.

From SQL Queries to Real-Time Conversations

The practical impact of this secure gateway was validated during a 90-day Pioneer Program that saw participation swell from four to 64 churches. Throughout the pilot, ACST pushed 19 production releases, rapidly incorporating user feedback to refine the system's capabilities across contacts, groups, attendance, giving, and communications.

Participants reported that MP-MCP enabled them to uncover complex relationships across disparate datasets—like giving, attendance, and group participation—that previously required running multiple reports and performing manual analysis. The primary benefit was a dramatic increase in speed and efficiency, particularly in data maintenance.

"Before MP-MCP, getting anything out of MinistryPlatform meant writing SQL or digging through reports, then translating it for whoever needed it," said Jason Johnston, Director of Engagement at Bent Tree in Carrollton, TX. "We wanted data work to move as fast as a conversation. With MP-MCP, cleanup work that used to get skipped because it wasn't worth the hours now just gets done.”

Johnston provided concrete examples of the tool's impact, noting his team has used it to correct over 1,500 mismatched records, build new groups, and fix data audit issues in a fraction of the time it would have taken previously. "The biggest difference is that I can catch issues and ask follow-up questions mid-analysis instead of after the fact," he added. This iterative, conversational approach leads to cleaner data and frees up valuable staff time for ministry-focused activities.

The Double-Edged Sword of Open Standards

Under the hood, MP-MCP is built on the Model Context Protocol (MCP), an open-source framework championed by AI leaders like Anthropic, OpenAI, and Google. Hailed as a "USB-C port for AI," MCP standardizes how AI models connect to external tools and data sources, solving a major interoperability challenge. It’s what allows MP-MCP to seamlessly connect with various AI assistants that support the standard.

However, security researchers have noted that while MCP is a powerful enabler, the protocol itself does not include inherent data privacy or security controls. Its power to connect LLMs directly to data sources also introduces significant risks, including prompt injection attacks, credential exposure, and the potential for an AI to be tricked into executing unauthorized commands. The responsibility for securing these connections falls squarely on the implementer.

This is where ACST's work becomes a valuable case study. Instead of simply adopting the open standard, the company has built a comprehensive security wrapper around it. The six layers of protection are not part of MCP; they are ACST's solution to the protocol's inherent risks. This demonstrates a mature approach to engineering: leveraging the power of an open ecosystem while meticulously building the safeguards that the standard itself omits, creating a product that is both functional and trustworthy.

Navigating the Governance Gap in Faith Tech

The launch of MP-MCP arrives at a critical moment for faith-based organizations. While interest in AI is high, so is anxiety. Recent studies show that while over 80% of church leaders are concerned about AI's impact on data privacy, fewer than 10% of churches have a formal AI policy in place. This creates a significant "governance gap," where crucial decisions about data handling are effectively outsourced to software vendors.

In this environment, a vendor's product design becomes the de facto policy for its users. ACST's decision to build a secure-by-default system—with disabled write access, PII masking, and strict permission controls—provides a strong governance foundation for churches that are just beginning to navigate the complexities of AI. By prioritizing data stewardship in its architecture, MP-MCP empowers ministry teams to experiment with AI's benefits without inadvertently compromising their ethical and pastoral responsibilities to their congregations.