No real esptool operation backs "optimize firmware binary". Size
optimization happens at compile-time via ESP-IDF menuconfig, not as
a post-processing step on a .bin file. The analysis use case is
already covered by esp_firmware_analyze.
The ROM bootloader (--no-stub) runs at ~62 kbit/s which causes
timeouts on large images. The stub flasher is dramatically faster
and works correctly over QEMU socket connections.
Replace stub implementations with real esptool subprocess calls.
Each operation runs esptool as an async subprocess with timeout,
matching the chip_control pattern. All four tools now work with
both physical ports and QEMU socket:// URIs.
Port allocator was counting stopped QEMU instances as occupying ports,
preventing new starts after stop-all. Only running instances now block
port reuse. Stopped instances with flash images on disk are preserved
for esp_qemu_flash; truly stale entries are purged.
Sort glob results when auto-detecting QEMU binaries to reliably
pick the newest version. Previously filesystem ordering could select
an older build (8.2.0 instead of 9.0.0), missing ESP32-S3 support.
Remove ESP32-S2 from CHIP_MACHINES — the Espressif QEMU fork has
no esp32s2 machine type.
Add QEMU section to README with install instructions and tool reference.
Enrich MCP tool docstrings with workflow context, cross-references to
related tools, and guidance on when to use each tool — these descriptions
are read by the calling LLM to decide tool selection.
Smoke tested against real Espressif QEMU fork — esptool successfully
identifies, reads flash_id, and writes firmware to emulated ESP32 via
socket://localhost:5555. Enriched CHIP_MACHINES with per-chip efuse
data, GPIO strap values, and watchdog timer driver names derived from
ESP-IDF's qemu_ext.py.
