From Curiosity to Capability: How Experimenting with the MCP Server Led Us to AI Admin Agents

When you work at a Salesforce consulting company, your day-to-day usually revolves around solving business problems, building clever automations, and occasionally wondering why that flow suddenly broke after the last release. But every now and then, a new piece of technology drops that changes the conversation entirely.

For us, that was the MCP (Model Context Protocol) server.

What is the MCP Server?

Think of MCP as a secure, universal bridge between AI models and your live systems — in our case, Salesforce orgs. It’s a protocol that gives AI safe, governed, and contextual access to things like metadata, data, and developer tooling, without exposing the entire kingdom to the dragons of uncontrolled access.

It’s not just an API — it’s a conversational interface layer. The MCP server handles the orchestration, context injection, and guardrails so that when an AI model is “inside” your environment, it behaves more like a trained assistant than a rogue script runner.

In simpler terms:

• Before MCP – AI could give you advice, but it didn’t really know your org.
• After MCP – AI can see just enough of your org to take meaningful action, run commands, and return results — all under strict governance.

How We Got Hooked

We started small. A simple experiment: could we use MCP to let an AI agent fetch schema details, run a query, and summarize results in natural language? The answer was yes — and it was fast.

Then we added a few more actions:

• Deploy a metadata component.
• Spin up a scratch org for testing.
• Compare two profiles and list permission differences.

Within a week, we realized we weren’t just running tests — we were building the foundations of AI-powered Salesforce admin agents.

What AI Admin Agents Can Do

Once we connected MCP to the right commands and contextual data, the possibilities multiplied:

• Automated User Management
  “Deactivate all inactive users who haven’t logged in for 90 days” → done in seconds, with a full audit trail.
• Org Documentation On Demand
  “Generate a permissions matrix for all profiles” → agent pulls it directly from Salesforce and formats it in a spreadsheet.
• Data Quality Sweeps
  “List all accounts missing a primary contact and auto-assign follow-up tasks” → no clicks, no SOQL wrangling.
• Metadata Changes with Guardrails
  “Add a validation rule for postal code formats” → AI drafts the rule, runs it in a sandbox, and shows the diff before you approve.
• Org Health Checks
  “Find all hard-coded IDs in validation rules and flows” → agent searches metadata, flags issues, and suggests fixes.

Why This Matters for Salesforce Teams

The MCP server means we can now:

• Scale admin tasks without scaling headcount.
• Enforce governance while still empowering AI to act.
• Bridge skill gaps by letting newer admins use plain language to run complex tasks.

For us, it’s not about replacing admins — it’s about removing the repetitive, high-friction work, so the human experts can focus on design, strategy, and stakeholder alignment.

What’s Next

We’re already testing multi-agent workflows — where one AI agent handles metadata analysis, another runs data fixes, and a third documents everything in Confluence. All orchestrated through MCP.

If the early results are anything to go by, we’re stepping into an era where “AI-powered org management” won’t just be a buzzword — it’ll be a standard practice. And for a Salesforce consulting team like ours, that’s an incredibly exciting frontier.

‍