Snippy

Created By

Azure-Samples8 months ago

🧩 Demo: Build AI-powered MCP Tools for GitHub Copilot using Azure Functions, OpenAI, AI Agents & Cosmos DB. Manages code snippets w/ vector search.

# mcp

# azure-functions

Overview Content Tools Comments

Content

Snippy · Intelligent Code-Snippet Service with MCP Tools

Snippy is an Azure Functions–based reference application that turns any function into an MCP (Model Context Protocol) tool consumable by GitHub Copilot Chat and other MCP‑aware clients. The sample implements a production‑style code‑snippet service with AI‑powered analysis:

Save snippets – persists code, metadata and OpenAI embeddings in Cosmos DB DiskANN
Semantic retrieve – vector search over embeddings
AI Agents – generate a deep wiki or language‑specific code style guide from stored snippets
Durable fan‑out/fan‑in with Blueprints – in experimental branch for large‑scale processing
Microsoft Fabric integration – in gk/fabric branch demonstrating how to build Agents with Fabric Data Agents

The project ships with reproducible azd infrastructure, so azd up will stand up the entire stack – Functions, Cosmos DB, Azure OpenAI and Azure AI Agents – in a single command.

Important Security Notice This repository is intended for learning and demonstration purposes. Do not deploy it to production without a thorough security review. At a minimum you should:

Swap connection strings for Managed Identity + Azure Key Vault

Restrict network access to Azure services via Private Endpoints or service‑tags

Enable GitHub secret‑scanning and CI security tools

Features • Architecture • Getting Started • Guidance

Features

Remote MCP trigger – expose Functions as real‑time SSE tools
AI‑assisted documentation – "deep‑wiki" and "code‑style" agents create rich Markdown (Mermaid, diagrams, etc.)
Vector search on Cosmos DB DiskANN – low‑latency semantic retrieval
One‑click deploy – azd up provisions and deploys code & infra
Codespaces & Dev Containers – fully configured dev environment in your browser or local VS Code

Tool Matrix

Tool Name	Purpose
`save_snippet`	Save code snippets with vector embeddings for semantic search
`get_snippet`	Retrieve previously saved code snippets by their unique name
`code_style`	Generate language-specific code style guides from saved snippets
`deep_wiki`	Create comprehensive wiki documentation by analyzing code snippets

Architecture Diagram

Snippy Architecture

flowchart LR
    %% ─── MCP Hosts & Clients (local) ──────────────────────────────
    subgraph mcphosts["MCP Hosts & Clients (Your Computer)"]
        Host["Host<br/>(VS Code / IDE)"]
        Client["Client<br/>(GitHub Copilot)"]
    end

    %% ─── Application on Azure (remote) ────────────────────────────
    subgraph app["Application (Azure)"]
        Snippy["MCP Server<br/>Snippy Triggers<br/>(Function App)"]:::dashed
        Foundry["Foundry Agent<br/>Deep Wiki · Code Style"]
        Cosmos["Cosmos DB<br/>Operational + Vector DB"]:::datasource
        AOAI["Azure OpenAI<br/>text‑embedding‑3‑small"]
        FabricDA["Fabric – Data Agent"]
        VectorTool["Tools – Vector Search"]
    end

    %% ─── Local interactions ───────────────────────────────────────
    Host <--> Client

    %% ─── MCP protocol to Azure ────────────────────────────────────
    Client <-- "MCP Protocol (SSE)" --> Snippy

    %% ─── Bindings & data flow inside Azure ────────────────────────
    Snippy -- Bindings --> AOAI
    Snippy --> Cosmos
    Snippy --> Foundry
    Foundry --> FabricDA
    Foundry --> VectorTool

    %% ─── Styling ──────────────────────────────────────────────────
    classDef datasource stroke-width:2,stroke-dasharray:5 5
    classDef dashed stroke-width:2,stroke-dasharray:5 5,fill:transparent
    class Cosmos datasource
    class Snippy dashed
    style mcphosts fill:transparent
    style app fill:transparent

Getting Started

You can run Snippy in GitHub Codespaces, VS Code Dev Containers, or your local environment. The fastest path is Codespaces.

Snippy requires an Azure region that supports text‑embedding‑3‑small (or a compatible embeddings model) and Azure AI Agents. The azd workflow prompts you for a region; we recommend eastus for best availability.

GitHub Codespaces

Click Open in Codespaces above (first badge) – the container build may take a few minutes.
When the terminal appears, sign in:
```
azd auth login --use-device-code
```
Launch the stack:
```
azd up
```
Once deployment completes, copy the printed MCP URL and open GitHub Copilot Chat → Agent mode to try commands like “Save this snippet as hello‑world”.

VS Code Dev Containers

Prerequisites: Docker Desktop + the Dev Containers extension.

Click the Dev Containers badge (second badge) or run Remote‑Containers: Open Repository in Container from VS Code.
Sign in and launch as shown for Codespaces:
```
azd auth login
azd up
```

Local Environment

Prerequisites

azd CLI
Python 3.11 + uv
Node 18+ (for Functions Core Tools)
Azure Functions Core Tools v4 (npm i -g azure-functions-core-tools@4 --unsafe-perm)

Quickstart

# 1. Clone & init
azd init --template Azure-Samples/snippy

# 2. Sign in
azd auth login

# 3. Provision & deploy
azd up

The CLI prints the Function App URL, MCP endpoint and system key when finished. To remove all resources later:

azd down --purge

Guidance

Region Availability

Azure OpenAI model support varies by region. Verify availability here and choose the same region for all Azure resources. eastus and swedencentral are good default choices.

Costs

Estimate monthly cost using the Azure Pricing Calculator. Major components:

Azure Functions – Consumption / Flex tiers
Cosmos DB – Serverless or provisioned throughput
Azure OpenAI – pay‑as‑you‑go per 1K tokens
Azure AI Agents – per‑execution billing (preview)

Security

Snippy uses User-Assigned Managed Identity for secure service-to-service authentication. The infrastructure is configured with:

User-Assigned Managed Identity on the Function App with appropriate RBAC roles:
- Cosmos DB Data Contributor
- Storage Blob Data Owner and Queue Data Contributor
- Application Insights Monitoring Metrics Publisher
- Azure AI Project Developer

For production deployments, we recommend: