TRACE32 MCP Server

An MCP server that wraps Lauterbach TRACE32's public remote API (RCL) via the official lauterbach-trace32-rcl (PyRCL) package. It lets an AI agent (e.g. GitHub Copilot agent mode) drive TRACE32 PowerView for three MVP use cases:

• Debugging — read/write memory, registers and variables; run control (go / break / step); breakpoints

• Resource inspection — connection/run state, CPU, memory & register snapshots, PRACTICE function evaluation

• Script execution & flash — run PRACTICE .cmm scripts, including SW target flash

Prerequisites

• Python 3.9+

• TRACE32 PowerView already running with the remote API (RCL) enabled. Add this to your config.t32 (blank line before/after each block):

  RCL=NETTCP
  PORT=20000

  Sanity-check it works before wiring up MCP:

  t32rem localhost PRINT VERSION.SOFTWARE()

Related MCP server: jlink-mcp

Install

pip install -e .

Configuration (environment variables)

Safety guards

All read and debug-control tools are always available. Destructive operations are OFF by default and must be explicitly enabled:

• T32_ALLOW_WRITE=true → memory/register writes, arbitrary command passthrough, script execution

• T32_ALLOW_FLASH=true → flash programming (independent of T32_ALLOW_WRITE)

When a guard is closed, the tool returns an error explaining which flag to set — it never silently touches the target.

Register with GitHub Copilot (VS Code)

Workspace file .vscode/mcp.json is included. Adjust the env values (host/port, and the guard flags when you intend to write/flash), then start the server from the MCP view or the MCP: List Servers command. You can also run it standalone:

python -m trace32_mcp

Tools

Tests

pytest

Unit tests mock PyRCL, so no running TRACE32 is required. For end-to-end checks, the TRACE32 Instruction Set Simulator can stand in for real hardware.

Roadmap (post-MVP)

Trace/CTS inspection, symbol/source browsing, multicore selection, call stack, peripheral register views, structured variable trees, multi-instance routing.