How MCP Is Changing Public Sector AI in Salesforce
Introduction
We’re at a pivotal moment in the evolution of enterprise AI — and if you're working with Salesforce in the public sector, this is something you’ll want to pay attention to.
The Model Context Protocol (MCP) might sound like yet another acronym in an already crowded space, but what it enables under the hood could reshape how we think about integrating AI into enterprise architecture — particularly in environments where interoperability, accountability, and compliance are non-negotiable.
What is MCP, Really?
Put simply, MCP is a universal interface for AI agents. Originally introduced by Anthropic and now adopted by OpenAI, Google, Microsoft, and—importantly for us—Salesforce, it provides a standardised way for AI models to talk to different systems, fetch real-time context, and take intelligent actions.
"Think of it as the Rosetta Stone for AI integrations."
MCP works through a clean separation of concerns:
- The Host is the environment where AI operates — could be your CRM, case management platform, or knowledge base.
- The Client is the protocol-aware layer that speaks MCP and routes requests.
- The Server is the back-end interface that executes these requests, often by querying data systems, triggering business logic, or talking to other agents.
For the public sector, where data sits in silos (and for good reason), this kind of loosely coupled, interoperable model is a big deal.
It means we can start wiring up intelligent workflows without needing brittle, one-off APIs for everything.
What This Means for Salesforce — and the Public Sector
Salesforce is already betting big on AI through Agentforce, its in-house platform for deploying task-specific AI agents across service, sales, and operations.
The idea is solid: embed agents directly in the platform where your data and users already live.
But here's the challenge. Salesforce’s AI today is still relatively siloed.
Agentforce agents mostly operate within the Salesforce environment, which is fine until you need them to pull data from, say, a legacy procurement system, a case file in SharePoint, or even your internal document repository.
That’s where MCP changes the game. With MCP-supported integration — now available in Beta through MuleSoft AnyPoint — Agentforce agents could start reaching beyond the
Salesforce data layer, tapping into the wider context: financial records, citizen engagement data, operational systems, or even third-party intelligence platforms.
This is especially critical for public sector use cases like:
- Coordinated case management across health, housing, and welfare
- Multi-agency incident response
- Cross-departmental reporting and compliance tracking
The Opportunity — and the Risk — for Salesforce
Here’s the strategic tension. Salesforce built Agentforce assuming customers would keep their AI strategy inside the Salesforce walled garden.
MCP, by contrast, levels the playing field — now any AI agent, regardless of vendor, can query your Salesforce org as just another MCP-compatible server.
This opens the door to much cheaper “overlay” solutions — an LLM powered by OpenAI or Gemini could interact with Salesforce data without needing Agentforce licenses at all.
That’s both a threat and an opportunity.
But Salesforce has two major advantages:
- Deep customer context. No other system knows your customer, citizen, or stakeholder like Salesforce does.
- Enterprise-grade trust. With the Einstein Trust Layer, Salesforce has built strong privacy and governance tooling that makes it viable in even the most sensitive public sector deployments.
So the challenge now is: how does Salesforce make it easy to extend Agentforce using MCP — without losing control of the experience?
Enter AgentExchange: A Marketplace for AI Agents
If Salesforce wants to win the MCP game, it needs to double down on usability and value — and a curated AgentExchange marketplace could be the answer.
Imagine this:
- Pre-built, plug-and-play agents tailored for public sector use cases: FOI response bots, citizen inquiry triage agents, grants eligibility checkers, etc.
- Powered by Salesforce but interfacing with everything else through MCP—Office 365, internal APIs, government data feeds.
- Governed by the same trust and security model as Salesforce core.
- Priced flexibly to encourage experimentation.
This would give technical teams a faster path to deploy agents that work across boundaries — without needing a two-year IT project to wire them up.
Final Thoughts from Digital Modus
From where we sit at Digital Modus, working with complex, often multi-agency public sector clients, this is a moment of real possibility.
But it’s going to take clarity from Salesforce — both in terms of technical standards and commercial models — to make MCP real in the field.
The recent flexibility in Agentforce pricing is a step in the right direction. Extending that to a thriving AgentExchange with clear rev share models could unlock real innovation at the edge.
This isn’t just a technical pivot — it’s a philosophical one.
We’re moving from “one platform to rule them all” to “intelligent agents that play well with others.”
Let’s build accordingly.
