Claude Code OTel logs now tag which tool made each call

Claude Code's OpenTelemetry export picked up three new fields in v2.1.214: message.uuid, client_request_id, and tool_source, according to the changelog. The one worth wiring up immediately is tool_source.

Before this, an OTel log event told you a tool ran, but not where it came from. If you had Claude Code hooked up to three MCP servers plus its built-in file and shell tools, every trace looked the same. Now tool_source is the join key: it tells you whether a call came from a built-in tool or from a specific MCP server. That's the difference between guessing and knowing which connector is burning your tokens, or which server got called right before an agent run went sideways.

The same release added a CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH environment variable to configure the existing 60 KB truncation on content attributes, so teams piping full tool payloads into their observability stack can raise or lower that ceiling instead of losing data to a hardcoded limit.

It also put reasoning effort into the subagentStatusLine payload. If you're running custom multi-agent setups, that means a status line row can now show which model and which effort level a given subagent is using, live, instead of you having to infer it from behavior.

None of this is a new capability so much as new visibility into behavior that was already happening. If you run Claude Code with MCP servers in production and you've been debugging token spend or failure chains by eyeballing logs, this is the attribute that turns that into a query. Worth checking whether your OTel collector config is already picking up the new fields or silently dropping them.