QIDI Studio MCP Server

MCP server for managing QIDI Studio 3D printing profiles locally.

Русская версия → README_RU.md

Connects AI assistants (Claude, Cursor, etc.) directly to your local QIDI Studio configuration files. Reads and writes filament, process, and machine presets via the Model Context Protocol.

⚠️ ALPHA VERSION — USE AT YOUR OWN RISK The author is not responsible for lost or corrupted presets. Always backup before using MCP.

Why this exists

Real-time printer control — start, pause, heat — is easier done by hand in QIDI Studio. MCP is not needed there.

Print settings, however, are a completely different story.

QIDI Studio contains hundreds of parameters: temperatures, retractions, fan speeds, support patterns, gaps, accelerations. Every material (PETG, PLA, ABS, composites) and every nozzle requires its own balance. Memorizing all of this and not missing a detail is difficult even for an experienced user.

This MCP solves exactly that: it helps AI help you pick, create, and save the right preset — safely, controllably, and with minimal token consumption. You still see the final settings in QIDI Studio before sending the job to print.

The core problem

• Too many settings across different materials and nozzles

• Easy to forget a key parameter (e.g., fan_max_speed for PETG or support_type for complex geometry)

• Manually browsing JSON profiles in folders is inconvenient and slow

• No centralized way to "try option A, compare with option B, save the best"

What the MCP does

• Auto-detects the current project from QIDI Studio configs — no need to type printer and nozzle manually

• Filters profiles by printer + nozzle + material — you see only what is relevant

• Safely creates presets — AI proposes, you confirm, a user preset is saved without overwriting system ones

• Validates — checks temperatures, enum values, and unknown keys before writing to disk

⚠️ Backup First

The server writes directly to QIDI Studio's configuration files. Mistakes happen.

Where your presets are stored:

Backup:

# macOS
cp -r ~/Library/Application\ Support/QIDIStudio/user/default ~/Desktop/QIDIStudio-backup

# Windows
xcopy "%APPDATA%\QIDIStudio\user\default" "%USERPROFILE%\Desktop\QIDIStudio-backup" /E /I

# Linux
cp -r ~/.config/QIDIStudio/user/default ~/QIDIStudio-backup

Restore:

# macOS
cp -r ~/Desktop/QIDIStudio-backup/* ~/Library/Application\ Support/QIDIStudio/user/default/

# Windows
xcopy "%USERPROFILE%\Desktop\QIDIStudio-backup" "%APPDATA%\QIDIStudio\user\default" /E /Y

# Linux
cp -r ~/QIDIStudio-backup/* ~/.config/QIDIStudio/user/default/

Quick Start

Requirements

• Python 3.10+

• QIDI Studio installed (the server reads its built-in profiles)

Install

git clone https://github.com/QIDITECH/QIDIStudio-MCP.git
cd QIDIStudio-MCP
pip install -r requirements.txt

Connect to Claude Desktop

Option 1: Automatic (uv)

uv run mcp install mcp_server.py --name "QIDI Studio"

Option 2: Manual

Edit Claude Desktop settings:

• macOS: ~/Library/Application Support/Claude/settings.json

• Windows: %APPDATA%\Claude\settings.json

{
  "mcpServers": {
    "qidi-studio": {
      "command": "python",
      "args": ["/path/to/QIDIStudio-MCP/mcp_server.py"]
    }
  }
}

Environment Variables

AI Agent Instructions (Required)

Step 1: Detect current project (auto)

get_current_project_info()  → {"machine": "Q2 0.4 nozzle", "filament": "Generic PETG...", "process": "0.20mm Standard @Q2"}

If a project is detected, propose using the detected printer instead of asking. Still confirm with the user.

Step 2: If no project detected, ask the user

Show numbered lists and ask to pick by number. Use compact=True and limit=10 to keep responses tiny:

Detected printers:
1. Q2
2. X-Plus 4
3. X-Max 4
Which printer? (enter number):

Nozzles for Q2:
1. 0.2
2. 0.4
3. 0.6
4. 0.8
Which nozzle? (enter number):

Filter semantics for process profiles: 0.20mm Standard @Q2 has no nozzle in its name. It matches printer_model="Q2" + nozzle_size="0.4" only if its compatible_printers contains "Q2 0.4 nozzle". Without compatible_printers or for a non-existent nozzle, the profile is filtered out. Always pass both printer and nozzle for process queries.

list_profiles response format: On success, an array of profiles is returned. On empty filtered result, an object {"profiles": [], "debug": {...}, "hint": "..."} is returned. Check the response type: array = results, object with profiles = empty + reasons.

Debug on empty results: The debug field contains total_available, sample_names, name_matches, fallback_matches, and excluded_examples — use this to understand why the filter dropped profiles.

Filter semantics for material_type: Only applies to filament profiles. For process, use printer_model instead.

Step 3: Always confirm before creating/updating

I will create preset "Glass PETG @Qidi Q2 0.4" with:
- Printer: Q2 (0.4 nozzle)
- Base: Generic PETG @Qidi Q2 0.4 nozzle
- Settings: nozzle_temperature=["255"], fan_max_speed=["0"]
Confirm? (yes/no)

NEVER do this

❌ list_profiles("filament", "all")          → returns 1000+ profiles
❌ list_profiles("filament", "all", material="PLA") → still hundreds across all printers

Correct token-efficient flow

get_current_project_info()                → auto-detect or skip
get_available_printers()                  → if not detected
get_available_nozzles("Q2")               → if not detected
get_available_vendors()                   → ["Generic", "QIDI", "HATCHBOX"]
get_available_materials("Q2", "0.4")     → ["PETG", "PLA"]
list_profiles("filament", "all", "Q2", "0.4", "PETG", 50, 0, True)  → 3 compact profiles
read_profile(...)                         → inspect one
// CONFIRM with user before:
create_preset(...)                        → create custom

MCP Tools

Discovery (call these first, always filtered)

Profile Operations

Token-Saving Design

This MCP is designed so that the AI agent spends minimum tokens on routine and maximum on decision-making.

• Settings are read from local JSON profiles directly — no hardcoded or outdated lists. discover_settings returns only what actually exists in the QIDI Studio profiles on your machine.

• Filtering cuts 90%+ noise — list_profiles with printer_model, nozzle_size, and material_type returns 3–15 profiles instead of 1000+.

• Compact mode — compact=True returns only name and source, no paths or metadata.

• Prefix filtering for settings — discover_settings("process", "support_", ...) returns dozens of fields instead of hundreds.

• Keys-only mode — keys_only=True gives a list of parameter names without values when you just need to know what is available.

• Pagination and limits — limit=50 and offset prevent accidental dumps of the entire catalog.

• Debug on empty results — if a filter returns [], the response includes debug with excluded_examples explaining why profiles were dropped. No guessing needed.

Examples

Create Glass PETG preset

get_available_printers()
get_available_nozzles("Q2")
get_available_vendors()
get_available_materials("Q2", "0.4")
list_profiles("filament", "all", "Q2", "0.4", "PETG", 50, 0, True)
create_preset("filament", "Glass PETG @Qidi Q2 0.4", "Generic PETG @Qidi Q2 0.4 nozzle",
  '{"nozzle_temperature": ["255"], "fan_max_speed": ["0"]}')

Vase mode process

get_available_printers()
get_available_nozzles("Q2")
get_available_process_qualities("Q2", "0.4")
list_profiles("process", "all", "Q2", "0.4", "", 10, 0, True)
clone_preset("process", "0.12mm High Quality @Q2", "0.12mm Vase @Q2", "system")
update_preset("process", "0.12mm Vase @Q2", '{"spiral_mode": "1", "wall_loops": "1"}')

Find support-related settings

discover_settings("process", "support_", False)
discover_settings("process", "", True)  # just key names

Testing

python test_server.py

Project Structure

QIDIStudio-MCP/
├── mcp_server.py           # MCP server (tools)
├── qidi_config_manager.py  # Local profile manager
├── test_server.py          # Test suite
├── requirements.txt        # Dependencies (mcp)
├── .gitignore
├── README.md               # This file (EN)
└── README_RU.md            # Russian version

Tested Platforms

Please open an issue if you test on Windows or Linux.

License

MIT