Power Rubber Duck MCP connector for Copilot Studio

April 20, 2026

High-stakes decisions usually break for the same reason: we only look from one angle.

Power Rubber Duck is a Power Platform custom connector that exposes a full MCP server to Copilot Studio. It gives your agent four analysis tools and a shared resource library, so you can challenge assumptions, compare tradeoffs, and check for reasoning blind spots before you commit.

Full source: GitHub repository

What this connector adds

You get two capabilities in one connector:

1. Tool-driven analysis for dynamic decision support
2. Resource-driven grounding for consistent frameworks and checklists

That combination matters. Tools help your agent reason in the moment. Resources keep the reasoning anchored to reusable standards.

MCP tools

MCP resources

The connector exposes 10 resources that Copilot Studio can read with resources/list and resources/read:

• 3 decision frameworks: investment, operational change, and strategic planning
• 2 knowledge assets: benchmarks and case studies
• 3 reasoning guides: bias checklist, decision process, and critical questions
• 2 best-practice guides: implementation and organizational decision-making

These resources make outcomes more consistent across prompts and users because the agent can pull from the same decision playbook every time.

How it works in Copilot Studio

User asks a decision question
  -> Copilot Studio reads a decision framework resource
  -> Copilot Studio calls get_second_opinion
  -> Copilot Studio calls analyze_risk
  -> Copilot Studio calls identify_cognitive_biases
  -> Copilot Studio synthesizes recommendation + rationale

The MCP endpoint supports:

• tools/list
• tools/call
• resources/list
• resources/read

It also includes direct REST operations for each tool and resource action, which is useful when you want to call them from Power Automate or test each operation independently.

Example decision flow

Say your team is deciding whether to enter a new market.

1. Read resource://decision-frameworks/strategic
2. Call get_second_opinion with analysis_depth: "deep"
3. Call analyze_risk for technical, financial, and organizational risks
4. Call identify_cognitive_biases on your current plan narrative
5. Call comparative_analysis across options: build, partner, acquire
6. Return a recommendation with risks, mitigations, and monitoring signals

That gives you a recommendation plus an audit trail of why the recommendation was made.

Deploy the connector

Prerequisites

• Power Platform CLI (pac)
• A Copilot Studio environment
• A Foundry-compatible model endpoint (local or cloud)

Validate and create

# Validate connector
ppcv ".\Power Rubber Duck"

# Create connector
pac connector create --api-definition-file ".\Power Rubber Duck\apiDefinition.swagger.json" `
                     --api-properties-file ".\Power Rubber Duck\apiProperties.json" `
                     --script-file ".\Power Rubber Duck\script.csx"

Configure model endpoint

In script.csx, set:

private string foundryEndpoint = "http://endpoint";
private string foundryModel = "phi-4";

You can also override these with connection parameters (foundry_endpoint and foundry_model) in the connector configuration.

Observability with Application Insights

The script includes telemetry hooks for MCP requests, tool calls, resource reads, model calls, and exceptions.

Once you set your instrumentation key in script.csx, you can track:

• Which tools are used most
• Which resources are read most
• Error patterns by operation
• Model response characteristics over time

This helps you improve both agent quality and runtime reliability.

Why this pattern is useful

Most agent decisions fail quietly. You get fluent output, but weak reasoning.

Power Rubber Duck pushes your agent to:

• Use multiple perspectives
• Challenge its own assumptions
• Compare options explicitly
• Ground recommendations in shared decision assets

If you care about decision quality, not just answer quality, this is a practical pattern to adopt.

Files in the project

Resources

• Power Rubber Duck source code
• Model Context Protocol
• Custom connectors overview
• Power Platform CLI overview
• Application Insights overview