mcode-monitor

mcode-monitor is a local usage monitoring panel built for MiniMax Code, delivered as a Mini App. This project has been accepted into the official Mini App repository; the current repository hosts the main development version, with all updates synced to the official repository. Its core purpose is to let users view token consumption for MiniMax Code. After installation, users can open the "Token Usage Dashboard" from the Mini App entry point in MiniMax Code to check usage data at any time. Key features: - KPI overview: token consumption, input/output/cache read metrics, cache hit rate, request count, average output speed - Token consumption time series: automatic time bucketing based on selected time range, with stacked bar charts for cache and input values paired with an output trend line - Model/session filtering: multi-select dropdowns, with sessions aligned to actual session files, applied in real time to all charts and tables - Per-model breakdown: one-click toggle to view output curves for each individual model - Model comparison: total input volume bar chart, paired with dual trend lines for call count and output speed - Token speed: per-request tok/s line chart for the most recent 60 requests - Recent call details and summaries: timestamp, model, session, token count, duration, speed - Persisted settings: filter conditions, time range, refresh interval, and theme are automatically saved - Light/dark dual themes that follow system settings - Time range options: last 1 hour, last 24 hours (default), last 7 days, last 30 days, all time - Auto-refresh options: 5s, 10s, 30s, manual refresh Installation requires Python 3.8+ on the local machine, with no pip-installed dependencies needed. Simply place the miniapps/mcode-usage-monitor folder from the repository into the ~/.minimax/plugins/ directory: on Windows this path is C:\Users\your-account-name\.minimax\plugins\mcode-usage-monitor, and on macOS/Linux it is ~/.minimax/plugins/mcode-usage-monitor. To uninstall, simply delete this folder. The tool reads the local database in read-only mode, so it does not interfere with active MiniMax Code sessions. Both English and Chinese documentation are included in the package. Metric definitions: Token consumption is calculated as input tokens plus cache read tokens plus output tokens; cache hit rate is the share of prompt content that directly hits the cache; output speed is the weighted average of output tokens divided by request duration. Billing and quota values are subject to the usage page displayed within the MiniMax Code product. Per the project notes, all code and documentation is generated and iterated on by MiniMax Code (the AI coding assistant), with the author responsible for requirement definition, direction decisions, and acceptance checks. The project concept is adapted from the zcode-monitor project. The third-party component ECharts (Apache License 2.0) is used, and is packaged locally to support offline use. The project is released under the MIT License.