This sample demonstrates how to use a local MCP (Model Context Protocol) client with a ChatClientAgent using the Responses API directly.
- Connecting to an MCP server via HTTP (Streamable HTTP transport)
- Resolving MCP tools locally and wrapping them with logging
- Using
DelegatingAIFunctionto add custom behavior to MCP tools - Passing locally-resolved MCP tools to
ChatClientAgent
- .NET 10 SDK or later
- Microsoft Foundry service endpoint and deployment configured
- An authenticated Azure identity (for example, sign in with
az login)
Set the following environment variables:
$env:FOUNDRY_PROJECT_ENDPOINT="https://your-foundry-service.services.ai.azure.com/api/projects/your-foundry-project"
$env:FOUNDRY_MODEL="gpt-5.4-mini"dotnet run