OTEL_LOG_TOOL_CONTENT with MCP tool output: what Sume logs could leak

With Claude Code 2.1.283 and OTEL_LOG_TOOL_CONTENT=1, MCP tool outputs are recorded in the tool.output span event of your OpenTelemetry traces. If you connect Sume, that means whatever a Sume tool returns can land in your telemetry backend, so decide first what you are willing to store there.

What does the Claude Code change say?

The changelog entry for 2.1.283, dated Sep 25, lists MCP tool outputs in the OTEL tool.output span event with OTEL_LOG_TOOL_CONTENT=1. The changelog line is short. Check it for the release you run.

Which Sume outputs are sensitive?

Sume's docs list unsafe log content: API keys, signed URLs, raw private media URLs, and excessive user content or transcripts. The authentication page says to treat signed upload and download URLs as temporary secrets.

Hosted MCP includes assets_download_url and assets_upload_url among its tools, so a session that calls them can produce such URLs in a tool result.

What should the agent report instead?

The playbook for paid creates says to prefer Sume public ids and media.sume.com URLs in agent reports, and not to paste signed URLs, OAuth tokens or API keys into chat logs. The same habit keeps tool output cleaner in a trace.

What if a secret already reached the backend?

Sume's docs say to rotate API keys if they appear in logs or chat history. Signed URLs are temporary, but treat them as secrets until they expire, and limit who can read the trace store. Leave OTEL_LOG_TOOL_CONTENT unset unless you need tool content. If you do enable it, review who can query the collector and how long spans are retained, since a tool result is then stored beside the rest of your trace data.

Which outputs are worth keeping?

Job ids, request ids and status strings are the useful parts of a Sume trace, and Sume's docs list them as safe. A trace that carries jobs_status results is enough to explain what happened without storing any media URL.

If you do need content for debugging, capture it in a short-lived environment and switch it off afterward. That keeps the default trace free of anything on Sume's unsafe list.

Does this affect other Claude Code settings?

The same 2.1.283 entry lists a gateway hint header, x-claude-code-prompt-id, behind CLAUDE_CODE_GATEWAY_HINT_HEADERS=1. It is separate from tool content. Read the changelog entry before enabling either, and confirm what your collector stores.