ghcp-hosted-agents

You can stream long-running GitHub Copilot hosted agents by deploying them with the Invocations protocol over SSE, which sends progressive events to the caller and avoids non-streaming gateway response limits.

Yes, you can deploy Foundry hosted agents using BYOK authentication by minting bearer tokens via DefaultAzureCredential for the https://ai.azure.com/.default scope, entirely bypassing the need for a GITHUB_TOKEN.

You configure Foundry hosted agents by carefully setting up the session with skill_directories and mcp_servers configuration parameters during initialization to ensure compatible skill loading and MCP tool integration.

The Invocations protocol is used to stream SSE events from GitHub Copilot SDK agents to the caller, enabling progressive discovery for long tool loops like multi-site scraping where streaming progress matters more than minimal latency.

Yes, setting FOUNDRY_PROJECT_ENDPOINT and AZURE_AI_MODEL_DEPLOYMENT_NAME is required to properly configure BYOK authentication and invoke the hosted Foundry agent by POSTing JSON with an input string to the Invocations SSE endpoint.

You should avoid SSE streaming for GitHub Copilot hosted agents when your workloads require minimal orchestration latency, as this approach is specifically optimized for slow tool loops where streaming progress is the priority.