What is Copilotperplexitymcp

CopilotPerplexityMCP is a lightweight Model Context Protocol (MCP) server built with TypeScript and Express. It is designed to stream a tool catalog via a single Server-Sent Events endpoint and provides functions like echo, greet, and perplexity.search for Microsoft Copilot Studio agents and Power Automate flows to access real-time, citation-backed answers from Perplexity AI.

Use cases

Use cases for CopilotPerplexityMCP include automating responses in customer service applications, enhancing productivity tools with AI capabilities, and integrating AI search functionalities into business processes.

How to use

To use CopilotPerplexityMCP, clone the repository from GitHub, validate it, and set up the required GitHub secrets for Azure deployment. After setting up, you can invoke the provided functions through Microsoft Copilot Studio agents or Power Automate flows.

Key features

Key features of CopilotPerplexityMCP include compatibility with Node.js 20 LTS, deployment on Azure App Service for Linux, integration with Microsoft Copilot Studio and Power Automate, and the ability to access Perplexity AI’s REST API for real-time responses.

Where to use

CopilotPerplexityMCP can be used in various domains including software development, automation workflows, and any application requiring real-time AI-driven responses and data retrieval.

Copilot + Perplexity MCP Server ✨

A lightweight Model Context Protocol (MCP) server—built with TypeScript + Express, deployable to an Azure Linux Web App—that streams its tool catalog via a single Server-Sent Events endpoint and exposes echo, greet, and perplexity.search functions so Microsoft Copilot Studio agents (through a minimal Swagger-based custom connector) and Power Automate flows can invoke them for real-time, citation-backed answers from Perplexity AI.

Compatibility Summary

No other cloud services or proprietary dependencies are required—the repo is cloud-agnostic beyond Azure Web App and uses only standard HTTP + SSE.

End-to-End Setup — High-Level Checklist

1. Clone and Validate the Repo

git clone https://github.com/ITSpecialist111/CopilotPerplexityMCP.git
cd CopilotPerplexityMCP

Push a throw-away commit or open a PR → GitHub Actions runs the “Azure Web App CI” workflow. Green check = code builds & lints.

2. Add Required GitHub Secrets

Repo → Settings → Secrets → Actions → New repository secret.

3. Prepare the Azure Web App

Create resources:

az group create -n rg-mcp-demo -l uksouth
az appservice plan create -g rg-mcp-demo -n asp-mcp-demo --sku B1 --is-linux
az webapp create -g rg-mcp-demo -p asp-mcp-demo -n perplexmcpcopilot --runtime "NODE|20-lts"

App Settings (Portal → Configuration → Application settings):

• PERPLEXITY_API_KEY – your Perplexity key
• PERPLEXITY_MODEL – optional default model

Enable Application logging so you can tail the Log Stream.

Once the Web App exists, the publish-profile secret lets GitHub Actions deploy on every push.

4. Push → Deploy → Smoke Test

git commit --allow-empty -m "trigger deploy"
git push origin main

Actions builds, publishes to Azure.

Verify:

curl -N https://<webapp>.azurewebsites.net/sse    # should emit "event: tools"

5. Set Up the Custom Connector

Power Platform → Solutions → + New → Custom connector → Import an OpenAPI file.

• Upload swagger/mcp_connector.yaml (in this repo).
• Security → No authentication.
• AI Plugin (preview)
  • Fill Name, Description, Contact email, Legal URL.
  • Toggle Enabled for generative actions → On.
  • Create connection when prompted (no creds).
• Test tab → call MCP SSE Stream → should return the tool list.

6. Enable the Plugin in Copilot Studio

• Open your Copilot agent in the Studio designer.
• ✨ Overview → toggle Generative actions → On.
• Click + Actions, pick Perplexity MCP Connector under custom connections/MCP, toggle Use by this agent → On.
• Publish the agent.

Chat prompt example: “Using your Perplexity search tool, what’s the capital of France?”

Azure Log Stream should show:

GET /sse
POST /invoke/perplexity.search

—indicating the agent discovered the tools and invoked the one it needed.