Geargrafx Debugging

TurboGrafx-16 / PC Engine Game Debugging with Geargrafx

Overview

Debug TurboGrafx-16, PC Engine, and SuperGrafx games using the Geargrafx emulator as an MCP server. Control execution (pause, step, breakpoints), inspect the HuC6280 CPU and hardware (HuC6270 VDC, HuC6260 VCE, HuC6202 VPC, PSG), read/write memory, disassemble code, trace instructions, and capture screenshots — all through MCP tool calls. Hardware documentation is available in the references/ directory.

MCP Server Prerequisite

IMPORTANT — Check before installing: Before attempting any installation or configuration, you MUST first verify if the Geargrafx MCP server is already connected in your current session. In the default mode, call debug_get_status directly. If Geargrafx was intentionally started with --mcp-router, call get_tool_info with {"name":"debug_get_status"}, then call execute_tool with {"name":"debug_get_status","arguments":{}}. A valid response from either workflow means the server is active and ready.

Only if neither workflow is available or the call fails, you need to help install and configure the Geargrafx MCP server:

Installing Geargrafx

Run the bundled install script (macOS/Linux):

bash scripts/install.sh

This installs Geargrafx via Homebrew on macOS or downloads the latest release on Linux. It prints the binary path on completion. You can also set INSTALL_DIR to control where the binary goes (default: ~/.local/bin).

Alternatively, download from GitHub Releases or install with brew install --cask drhelius/geardome/geargrafx on macOS.

Connecting as MCP Server

Configure your AI client to run Geargrafx as an MCP server via STDIO transport. Example for Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "geargrafx": {
      "command": "/path/to/geargrafx",
      "args": ["--mcp-stdio"]
    }
  }
}

Replace /path/to/geargrafx with the actual binary path from the install script. Add --headless before --mcp-stdio on headless machines.

Hardware Documentation (References)

PC Engine hardware documentation is available in the references/ directory. Load them into your context when investigating specific hardware.

Debugging Workflow

1. Load and Orient

load_media → get_media_info → get_huc6280_status → get_screenshot

Start every session by loading the ROM, confirming it loaded correctly, then checking CPU state and taking a screenshot to understand the current game state. If a .sym, .lbl, or .noi file exists alongside the ROM, symbols are loaded automatically.

Load additional symbols with load_symbols or add individual labels with add_symbol.

2. Pause and Inspect

Always call debug_pause before inspecting state. While paused:

• CPU state: get_huc6280_status — registers A, X, Y, S, P (flags), PC, MPR mapping, timer, interrupts, I/O port, speed mode
• Disassembly: get_disassembly with a start/end address range — only shows executed code paths
• Call stack: get_call_stack — current subroutine hierarchy
• Memory: read_memory with a memory area tab ID (use list_memory_areas to discover available areas and their IDs)

3. Set Breakpoints

Use breakpoints to stop execution at points of interest:

Breakpoints support 5 memory areas: rom_ram (default), vram, palette, huc6270_reg, huc6260_reg.

Important: Read/write breakpoints stop with PC at the instruction after the memory access.

Manage breakpoints with list_breakpoints, remove_breakpoint.

4. Step Through Code

After hitting a breakpoint or pausing:

After each step, call get_huc6280_status and get_disassembly to see where you are.

5. Trace Execution

The trace logger records CPU instructions interleaved with hardware events (VDC, VCE, PSG, timer, CD-ROM, SCSI, ADPCM, input).

1. set_trace_log with enabled: true to start recording (optionally filter event types)
2. Let the game run or step through code
3. set_trace_log with enabled: false to stop (entries are preserved)
4. get_trace_log to read recorded entries

Available trace event filters: cpu_irq, vdc, vce, psg, timer, input, cdrom, adpcm, scsi. CPU tracing is always on.

Tracing is essential for understanding timing-sensitive code, interrupt handlers, and hardware interaction sequences.

Hardware Inspection

HuC6270 VDC (Video Display Controller)

• get_huc6270_status — VDC state: position, control, interrupt flags
• get_huc6270_registers — all 20 VDC registers (0x00-0x13), Address Register (AR), Status Register (SR)
• write_huc6270_register — write to a VDC register (0-19) or AR (20). Use vdc parameter (1 or 2) for SuperGrafx
• list_sprites — all 64 sprites with position, size, pattern, palette
• get_sprite_image — get a sprite as a PNG image

HuC6260 VCE (Video Color Encoder)

• get_huc6260_status — VCE state: position, sync signals, dot clock, control

HuC6202 VPC (SuperGrafx only)

• get_huc6202_status — VPC state: window settings, priority configuration

PSG (Programmable Sound Generator)

• get_psg_status — all 6 PSG channels: waveform, frequency, volume, noise, LFO, DDA

CD-ROM Subsystems (CD games only)

• get_cdrom_status — CD-ROM drive status
• get_cdrom_audio_status — CD audio playback status
• get_adpcm_status — ADPCM audio status
• get_arcade_card_status — Arcade Card status

CPU

• get_huc6280_status — full CPU state: registers, MPR, timer, interrupts, I/O, speed
• write_huc6280_register — modify a register live

Screenshots

• get_screenshot — current rendered frame as PNG

Use screenshots after stepping or continuing to see the visual impact of changes.

Common Debugging Scenarios

Finding an Interrupt Handler

1. Set an execution breakpoint at the IRQ vector address (read the Memory Map resource for the vector table)
2. debug_continue to run until the IRQ fires
3. get_huc6280_status + get_disassembly to see the handler code
4. get_call_stack to see how deep you are
5. add_symbol to label the handler address and any subroutines it calls

Diagnosing Graphics Corruption

1. debug_pause → get_huc6270_registers — check BAT, sprite attributes, scroll, DMA settings
2. get_huc6260_status — verify dot clock, color mode
3. get_screenshot — capture the current visual state
4. read_memory on VRAM and SAT areas to inspect tile/sprite data
5. Set read/write breakpoints (memory_area: vram or huc6270_reg) on display buffer addresses to catch corruption source

Analyzing a Subroutine

1. set_breakpoint at the subroutine entry point
2. debug_continue → when hit, get_huc6280_status
3. Step through with debug_step_into / debug_step_over
4. After each step: check registers, read relevant memory
5. add_symbol for the routine and any called subroutines
6. add_disassembler_bookmark to mark interesting locations

Tracking a Variable

1. add_memory_watch on the variable's address — watches are visible in the emulator GUI
2. Set a write breakpoint with set_breakpoint (type: write) on that address
3. When hit, get_disassembly reveals what code is modifying it
4. get_call_stack shows the call chain leading to the write

Timing Analysis

1. set_trace_log with enabled: true to start recording timer and VDC events
2. Let the game run through the section of interest
3. get_trace_log to see the interleaved CPU + hardware events
4. Check VDC interrupt timing via get_huc6270_status
5. Correlate timer fires with code execution in the trace

CD-ROM Game Debugging

1. load_media with a .cue file to load a CD-ROM image
2. get_cdrom_status to verify drive state
3. set_trace_log with cdrom: true and scsi: true to trace CD access patterns
4. get_adpcm_status to inspect ADPCM audio state
5. Set breakpoints on CD-ROM RAM areas to catch data loading

Memory Areas

Use list_memory_areas to get the full list with IDs and sizes. Common areas:

Organizing Your Debug Session

• Symbols: Use add_symbol liberally to label addresses you've identified — makes disassembly readable
• Bookmarks: Use add_disassembler_bookmark for code locations and add_memory_bookmark for data regions
• Watches: Use add_memory_watch for variables you're tracking across steps
• Save states: Use save_state / load_state to snapshot and restore emulator state at interesting points
• Rewind: Use get_rewind_status + rewind_seek to scrub back through recent execution history without manual save states
• Screenshots: Capture visual state with get_screenshot after significant changes

Rewind (Time Travel Debugging)

The emulator continuously records snapshots into a ring buffer during gameplay. You can seek to any recorded snapshot to restore full emulator state at that point in time — like time travel debugging.

Workflow

1. Check availability: get_rewind_status — returns snapshot count, capacity, buffered seconds
2. Pause: debug_pause — the emulator must be paused before seeking
3. Seek: rewind_seek with a snapshot number (1 = oldest, snapshot_count = newest)
4. Inspect: get_huc6280_status, get_disassembly, get_screenshot, read_memory, etc.
5. Iterate: Seek to different snapshots to narrow down when a bug first appeared
6. Resume or continue debugging: debug_continue to resume from the seeked state

Tools

Key Details

• Non-destructive seeking: rewind_seek loads a snapshot without removing it. You can seek to the same snapshot multiple times, or jump between different snapshots freely.
• Snapshot numbering: Snapshot 1 is the oldest available, snapshot_count is the newest (most recent).
• Buffer size: Configured by the user (default: 10 seconds). When full, oldest snapshots are overwritten.
• Granularity: Snapshots are taken every N frames (configurable). Default is every frame for maximum precision.

Bug Reproduction with Rewind

1. Let the game run past the bug occurrence
2. debug_pause → get_rewind_status to see how far back you can go
3. Binary search with rewind_seek: try the midpoint, check if the bug is visible (get_screenshot), then narrow the range
4. Once you find the exact snapshot where the bug appears, inspect CPU/memory state
5. Set breakpoints at the relevant code, then rewind_seek to a snapshot just before the bug and debug_continue