Pega Agentic AI with MCP: Orchestrating Multi Agent Intelligence the Enterprise Way

[RameshSangili]

Pega GenAI Cookbook - Recipes series

Enjoyed this article? See more similar articles in Pega Cookbook - Recipes series.

Pega Agentic AI with MCP: Orchestrating Multi‑Agent Intelligence the Enterprise Way

Enterprises are rapidly moving beyond isolated AI assistants toward agentic, outcome‑driven AI—where multiple specialized agents collaborate to complete real business work.

At the center of this evolution is Pega Agentic AI.

In this demo, Pega acts as the agentic brain and execution engine, owning the conversation, reasoning over context, and dynamically invoking external intelligence through the Model Context Protocol (MCP)—while maintaining enterprise‑grade governance, explainability, and control.

From the user’s perspective, this feels like one intelligent Pega agent.
Behind the scenes, multiple AI agents collaborate—fully orchestrated and governed by Pega.

---

Agentic Execution Model: One Call, Multiple Agents

From Pega’s perspective:

• One MCP invocation

• One conversation

• One case

• One governance model

Although multiple AI agents are involved, Pega makes a single decision and a single call.
The MCP server coordinates agent execution internally, while Pega remains the single system of interaction and control.

---

Combined Agentic Use Case: Loan Eligibility + Risk & Compliance Assessment

Both use cases—loan eligibility analysis and financial risk/compliance assessment—are executed through one Pega‑initiated MCP call.

Demo:

https://players.brightcove.net/1519050010001/default_default/index.html?videoId=6390965454112

Conversation Flow (Pega‑Centric)

1. A user engages a Pega Case Agent via natural language

2. Pega collects the full business context, including:

   • Credit score

   • Annual income

   • Requested loan amount

   • Customer background

   • Source of wealth

   • Transaction patterns

   • Risk flags

3. Pega reasons over the context and determines that:

   • Financial reasoning and

   • Risk & compliance analysis
     are both required

4. Pega invokes the MCP connector once,

One call from Pega
One conversational turn
One decision context

Agentic Technical Flow (Runtime Behavior)

Here’s how the architecture behaves during live execution:

1. Pega opens an SSE connection to the MCP server (/mcp)

2. MCP returns a session endpoint

3. Pega initializes shared context

4. Pega discovers available tools (tools/list)

5. Pega invokes MCP once using tools/call

6. MCP triggers:

   • Gemini → loan and financial analysis

   • Claude → risk and compliance analysis

7. Results stream back via SSE as a single response

8. Pega interprets the combined outcome

9. Pega applies policies, decisioning, and next‑best actions

10. Pega continues the conversation seamlessly

What This Enables for Real Enterprises

This pattern unlocks powerful, production‑ready outcomes:

• AI‑assisted underwriting

• Credit eligibility assessment

• Risk scoring and AML/KYC reasoning

• Fraud investigation insights

• Context‑aware recommendations

• Agent‑to‑agent workflows

All delivered through:

One conversation
One case
One governance model

Why This Matters

MCP enables intelligent collaboration—but Pega makes it enterprise‑ready.

With Pega Agentic AI:

• Pega owns the conversation

• Pega reasons over intent and context

• Pega decides which intelligence to invoke

• Pega governs execution and outcomes

^[1]

---

1. Footnotes ↩︎