MDX Technical Brief

Modern vehicle validation benches and track testing record hundreds of synchronous and asynchronous channels across powertrain, chassis, and ADAS networks. The resulting high-bandwidth bus logs routinely reach multi-gigabyte file sizes per test session.

Legacy calibration viewers struggle under this data volume because they attempt to load entire measurement arrays into desktop system memory, leading to interface freezes. Furthermore, cross-run comparison is hindered by timestamp drift across disparate logging instruments, while automation remains locked behind proprietary desktop macros.

MDX addresses large file performance through zero-copy memory mapping. Rather than buffering complete time-series arrays into RAM, the engine indexes raw MDF4 block headers (ID, HD, DG, CG, and CN blocks) directly from filesystem storage.

Waveform visualization uses the Largest-Triangle-Three-Buckets (LTTB) decimation algorithm. This preserves critical signal extrema, transient spikes, and zero-crossings when rendering millions of high-frequency samples on fixed-resolution displays. For empirical map generation, the computation core provides Kriging, radial basis functions (RBF), and spline surface interpolation.

To bridge automotive measurement with AI assistants, MDX embeds a native Model Context Protocol server operating over standard input/output (stdio) and Server-Sent Events (SSE). Any client adhering to the MCP standard connects with a single configuration block.

The server exposes twelve purpose-built tools: server health verification (mdx_health), session log enumeration (mdx_list_logs), binary mounting (mdx_load_log), channel schema discovery (mdx_query_channels), windowed statistical aggregation (mdx_signal_stats), decimated vector extraction (mdx_plot_signal), threshold transition detection (mdx_detect_events), cross-correlation pattern search (mdx_find_similar_events), and legacy workspace import (mdx_import_config).

The primary engineering view operates as an oscilloscope-grade workbench. Each signal lane features dedicated vertical range scaling, preventing high-voltage battery sweeps from compressing low-amplitude sensor signals. Dual time cursors calculate interval durations, scalar deltas, and first-derivative slopes (dV/dt).

For derived analysis, a server-side calculation engine computes virtual channels on the fly without modifying source binaries. Supported operations include physical accumulators (energy consumption, fuel mass), rolling statistical windows, fast Fourier transform (FFT) harmonic analysis, and Brake-Specific Fuel Consumption (BSFC) calculations.

MDX reads standard ASAM MDF v3 (.mdf) and MDF v4 (.mf4) files with full support for uncompressed and zlib-compressed data blocks. For network bus tracing, it decodes Vector BLF (.blf) and ASCII (.asc) logs across CAN, CAN FD, and LIN channels using CAN DBC (.dbc) and AUTOSAR ARXML (.arxml) network definitions.

Programmatic batch operations and headless pipeline integration are supported through a local REST service running on loopback (127.0.0.1:5001) and a companion Python SDK (mdx_client). Decimated plot arrays and tabular metrics can be exported directly to CSV or Excel.

Packaged with Tauri as a native desktop application for macOS, Windows, and Linux, MDX bundles its analytics backend locally. The software operates entirely without a floating licence server daemon, enabling continuous operation in air-gapped engineering facilities and test cells.

All telemetry processing occurs on local loopback. When AI agents query MDX over MCP, they receive only structured JSON summaries and downsampled plot vectors; proprietary raw measurement binaries are never uploaded or transmitted across external network boundaries.

MDX is an offline telemetry analysis and inspection platform. It is not physical bus interface hardware and does not replace hardware CAN transceivers (such as Vector VN hardware or Kvaser interfaces). It does not flash or write parameters directly to physical ECUs in a vehicle.

Furthermore, while AI agents can flag threshold anomalies and identify matching event intervals via MCP, MDX does not certify calibration sign-off without human engineering review. It does not provide multi-fleet centralized cloud storage unless explicitly connected to an external repository.