AI 程式開發工具
GitHub Copilot Opens Agent Telemetry and Token Limits in JetBrains, While BYOK Endpoints Gain Centralized Model Controls
GitHub has updated its JetBrains plugin to let agent workflows export OpenTelemetry data and to support input and output token limits for bring-your-own-key configurations and custom endpoints. Claude agent flows can also connect directly to MCP servers and custom agents, but deployment teams still need to validate tool-call content, cost attribution, and data redaction.
On July 27, GitHub updated Copilot in JetBrains IDEs, shifting the focus from expanding model options to operational controls for agents. The new version lets administrators configure OpenTelemetry exports in the plugin settings, allowing agent workflow execution data to flow into existing observability systems. BYOK and custom inference endpoints now support default `maxInputToken` and `maxOutputToken` values, while all built-in Copilot models can be enabled or disabled at once. For teams using both hosted models and internal gateways, these settings provide clearer boundaries for cost and model governance.
The update also brings MCP servers and custom agents into the Claude agent flow. Developers can apply project tools, team instructions, and specialized agents within the same IDE session without moving the workflow to an external terminal. Copilot CLI sessions now support forking, the `/rubber-duck` command, and to-do lists. MCP diagnostics, the model selector, user-prompt notifications, and path case handling for working sets have also been improved. These changes build on JetBrains' integration of Copilot into its native agent picker, while authentication continues to use a separate GitHub OAuth flow and a Copilot subscription.
The most important technical question is whether OpenTelemetry provides enough information for agent debugging. The announcement confirms that exports can be configured, but does not specify the complete span schema, redaction rules for prompts and tool parameters, trace correlation across forks, or differences in token counting among custom endpoints. Token limits also define only per-request boundaries and cannot guarantee the total cost of an agent session, since retries, branches, and MCP tools may continue to accumulate usage. Teams preparing to adopt the update should first inspect exported fields and sensitive data in a test project, then verify whether a fallback path remains after built-in models are disabled. They should also manage model, tool, and endpoint allowlists through version-controlled configuration instead of relying solely on individual developers' IDE settings.