Skip to content

README.md

Latest commit

 

History

History
105 lines (72 loc) · 5.86 KB

README.md

File metadata and controls

105 lines (72 loc) · 5.86 KB

Agent API (Azure Functions + LangChain.js)

Open project in GitHub Codespaces Node version TypeScript LangChain.js

Overview • API Endpoints • Development

Overview

The Agent API hosts the LangChain.js powered burger ordering agent. It:

  • Streams chat completions with intermediate tool + LLM steps
  • Connects to the Burger MCP server (Streamable HTTP) to invoke tools
  • Persists chat history + session titles in Azure Cosmos DB for NoSQL
  • Derives user identity from Azure Static Web Apps authentication
  • Emits OpenTelemetry traces (to Azure Monitor if configured, else local OTLP exporter)
Architecture

API Endpoints

Method Path Description
GET /api/me Returns (and lazily creates) the internal hashed user id for the authenticated user
GET /api/chats Lists all chat sessions for the current user
GET /api/chats/{sessionId} Returns messages for a specific session
DELETE /api/chats/{sessionId} Deletes a chat session and its messages
POST /api/chats/stream Streams an agent response for provided messages; creates/updates session

Streaming format

POST /api/chats/stream returns application/x-ndjson. Each line is a JSON object shaped like:

{
	"delta": {
    "content": "<partial text>",
    "role": "assistant",
    "context": { "currentStep": { ... } }
  },
	"context": { "sessionId": "<uuid>" }
}

Tool and LLM steps surface in context.currentStep / context.intermediateSteps enabling progressive UI rendering.

Development

Getting started

Follow the instructions here to set up the development environment for the entire Burger MCP Agents project.

Run the application

Use the following command to run the application locally:

npm start

This command will start the Azure Functions application locally. You can test the endpoints by opening the file api.http and click on Send Request to test the endpoints.

The agent API needs the Burger MCP server (and Burger API if running everything locally) to be running as well. You can start all the services at once by running npm start in the root of the project.

Note

If you have not deployed the Azure resources, it will fall back to in-memory data. You can test the API without deploying it to Azure.

Available Scripts

Script Description
npm start Start the development server with hot reload
npm run build Build the TypeScript source
npm run clean Clean build artifacts
update:local-settings Create or update local.settings.json needed by the Functions runtime

Configuration

The application uses environment variables for configuration:

Variable Required Purpose Default / Fallback
AZURE_OPENAI_API_ENDPOINT Yes Azure OpenAI endpoint used for chat completions –
AZURE_OPENAI_MODEL No Model name passed to LangChain.js gpt-5-mini
BURGER_MCP_URL Yes* Streamable HTTP endpoint of Burger MCP server http://localhost:3000/mcp
AZURE_COSMOSDB_NOSQL_ENDPOINT No Enables persistent chat + session titles In‑memory fallback
APPLICATIONINSIGHTS_CONNECTION_STRING No Sends traces to Azure Monitor Local OTLP exporter

*The code defaults BURGER_MCP_URL to the local dev port; production must set this.

Local OpenTelemetry traces

When running locally, spans are exported to http://localhost:4318/v1/traces and you can capture traces by running a local collector.

For example, you can use the AI Toolkit VSCode extension which includes a local OpenTelemetry collector.

After installing the extension, open the I Toolkit panel in the sidebar, go to the Tracing tool under Agent and workflow tools, and select Start Collector. When calling the agent API, you should see traces appear in the panel. You will then be able to inspect each trace in detail: The one named LangGraph contains the full sequence of LLM calls and tool invocations, allowing you to see how the agent arrived at its final response.