CopilotPerplexityMCP

# 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 | Layer | Compatible Targets | |-------|-------------------| | Runtime | Node.js 20 LTS (tested); works with any Node 18+ environment | | Hosting | Azure App Service for Linux (Basic B1 or higher) — container runtime uses the built-in NODE | | CI/CD | GitHub Actions (Ubuntu runners) via `azure/webapps-deploy@v3` | | MCP client | Microsoft Copilot Studio agents (Generative actions / "AI Plugin" preview)<br>Power Automate / Logic Apps (custom connector actions) | | LLM back-end | Perplexity AI REST API — any current public model (default `sonar`, but overridable) | | Local dev / testing | macOS, Windows, WSL2, or Linux with Node 20 + npm 10; cURL or Postman for raw SSE tests | 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 ```bash 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 | Secret | Value (where to get it) | |--------|-------------------------| | AZURE_PUBLISH_PROFILE | Download the Publish profile XML from the target Azure Web App → paste entire XML text. | | AZURE_WEBAPP_NAME | The exact Web App name (e.g. `perplexmcpcopilot`). | Repo → Settings → Secrets → Actions → New repository secret. ### 3. Prepare the Azure Web App **Create resources:** ```bash 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 ```bash git commit --allow-empty -m "trigger deploy" git push origin main ``` Actions builds, publishes to Azure. **Verify:** ```bash 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. <p align="center">   <a href="https://www.buymeacoffee.com/ITSpecialist" target="_blank">     <img src="https://img.shields.io/badge/Buy&nbsp;me&nbsp;a&nbsp;coffee-Support&nbsp;Dev-yellow?style=for-the-badge&logo=buy-me-a-coffee" alt="Buy Me A Coffee">   </a> </p>