[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f2r6s71aj0hwv":3},{"code":4,"message":5,"data":6,"timestamp":29},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":16,"summary_md":17,"topic":18,"github_topics":21,"license":25,"homepage":26,"updated_at":27,"history":28},"78\u002Fxiaozhi-esp32",2,"An MCP-based chatbot | 一个基于MCP的聊天机器人","C++",31875,7459,1729,784,[],"Firmware for ESP32 voice-assistant devices that connects cheap hardware to cloud LLMs, wake word included.","## The problem\n\nBuilding a voice assistant on a microcontroller normally means assembling pieces yourself: wake-word detection, Opus audio streaming over a stable transport, ASR\u002FLLM\u002FTTS orchestration, and some way for the model to actually touch the hardware. Most hobby projects stall somewhere in that pipeline. This project supplies a complete ESP32 firmware that turns a supported board into a voice terminal, with device-side MCP commands letting the model control speaker, LED, servo and GPIO, and cloud-side MCP extending it to smart-home or PC control.\n\n## What's different\n\nCompared to rolling your own on top of ESP-SR and a WebSocket client, the value here is breadth and integration: 138 board directories and 171 release variants, multiple transports (WebSocket, MQTT+UDP), offline wake word with custom words, and both conventional ASR+LLM+TTS and realtime end-to-end voice models. Compared to commercial voice-kit SDKs from chip vendors, it is hardware-agnostic across ESP32\u002FC3\u002FC5\u002FC6\u002FS3\u002FP4 and MIT-licensed for commercial use. The MCP layer is the distinguishing idea: treating the device as an MCP server rather than hard-coding each peripheral's commands.\n\n## Who it's for\n\nHardware hobbyists and small teams prototyping AI voice devices — desktop companions, pendants, robot dogs — who want working firmware without building the audio stack. It is also a reasonable reference for firmware engineers studying multi-board abstraction. Look elsewhere if you need a production-ready device cloud with guaranteed SLAs, or if your MCU is not ESP32: the codebase is deeply tied to ESP-IDF and its component ecosystem.\n\n## Cost of adoption\n\nThe toolchain requirement is steep: ESP-IDF v6.0.1 minimum, release\u002Fv6.1 recommended, with 5.x explicitly unsupported, and a full submodule\u002Ffullclean rebuild cycle after SDK changes. Practical use usually means registering on the official xiaozhi.me service for the default backend, or standing up one of the community servers (Python, Java, Go) yourself. Flashing prebuilt firmware is beginner-friendly; modifying it is not. The project is two years old and maintained actively, but the constant board-matrix growth means pin and IDF compatibility edge cases are a real support burden.\n\n## Verdict\n\nThis is an established project, not a spike: roughly 31.9k stars and the #2 spot on the monthly C++ board with +1,729 stars this month show interest still climbing. The churn is in hardware support, not core design, which is a healthy sign for a firmware base. Adopt it if you are building on ESP32 and want the voice pipeline solved; treat it as an SDK you must track rather than a drop-in binary. The main risk is coupling to the vendor's toolchain pace and to a default hosted backend.",{"slug":19,"name":20},"hardware","Hardware & IoT",[22,23,24],"chatbot","esp32","mcp","MIT","https:\u002F\u002Fxiaozhi.me","2026-10-24T17:36:37Z",[],1792871828]