What
When Claude Code runs on Anthropic's cloud machines, it can now report more about how its process starts and uses memory.
- Result messages, the summary sent at the end of a run, can carry
process_start_tuning,process_start_purge_delay_msandprocess_garbage_collections. - Garbage-collection counts, how often memory is cleaned up, are reported only when
CLAUDE_CODE_COUNT_GARBAGE_COLLECTIONS=1was set at process start. They come only from Anthropic-hosted cloud workers with telemetry on, and never from a process hosting other sessions. - The tuning value records which Bun and mimalloc variables were set, never their values.
- For
CLAUDE_CODE_REMOTEsessions, CPU, elapsed time, page faults and storage reads are sampled between startup milestones (spare loaded, claim, input ready, first message, first turn) and sent asprocess_stretches. Per-phase startup timings are also reported.
Why
These are diagnostics for Anthropic's cloud workers. They do not change how Claude Code behaves on your own computer, and the garbage-collection count stays off unless that variable is set.
Names in the bundleCLAUDE_CODE_REMOTE
Documented inclaude-code/agent-sdk/typescript
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
Confirmed since
Anthropic's documentation has since written up CLAUDE_CODE_REMOTE, on Customize sessions in self-hosted environments.
| `CLAUDE_CODE_REMOTE_MCP_CONFIG` | Path to a file in `CLAUDE_CONFIG_DIR` holding the control plane's delivered MCP servers. If your wrapper rebuilds the child's environment, carry this variable, [`CLAUDE_CODE_REMOTE`](https://code.claude.…self-hosted-environments-configuration see the edit
One source agreesOne thing we can check says the same as this entry.
Anthropic's documentation agrees
Anthropic's documentation has since written up CLAUDE_CODE_REMOTE, on Customize sessions in self-hosted environments.