# TrustMark Detection Troubleshooting Guide
## Issue: TrustMark Watermark Not Detected in Screenshot
### Quick Diagnosis
If your MCP server is not finding TrustMark watermarks in screenshots, follow these steps:
## Step 1: Test TrustMark Detection Directly
Use the diagnostic test script:
```bash
cd /Users/nhurwitz/mcp-content-credentials
node test-trustmark.js /path/to/your/screenshot.png
```
This will show you:
- Whether TrustMark detection is working
- Any error messages from the Python decoder
- The actual watermark data if found
## Step 2: Common Issues and Solutions
### Issue A: Screenshot Format Problems
**Problem:** TrustMark watermarks may not survive all screenshot methods.
**Check:**
- What tool did you use to take the screenshot?
- What format is the screenshot? (PNG, JPEG, etc.)
- Was the screenshot taken of the full image or cropped?
**Solutions:**
1. Save the watermarked image directly (File > Save) instead of screenshotting
2. If you must screenshot, use PNG format (lossless)
3. Ensure the screenshot captures the full image at 100% zoom
### Issue B: TrustMark Model Variant
**Problem:** Wrong model variant specified when detecting.
**Current Setting:** Model type 'P' (production/standard - default)
The server now uses the **'P' variant** by default, which is the standard TrustMark decoder variant. This matches the command-line usage:
```bash
trustmark decode --variant P -i [file_path]
```
**If you need to try the alternative 'Q' variant:**
Edit `src/c2pa-service.ts` line 25:
```typescript
private trustMarkService = createTrustMarkService('Q') // Try 'Q' variant if needed
```
Then rebuild: `npm run build`
### Issue C: Python/TrustMark Not Properly Installed
**Check Installation:**
```bash
python3 -c "from trustmark import TrustMark; from PIL import Image; print('✅ All dependencies installed')"
```
**Reinstall if needed:**
```bash
pip3 install --upgrade trustmark Pillow
```
### Issue D: Image Modification During Screenshot
**Problem:** OS screenshot tools may apply compression or processing.
**TrustMark is sensitive to:**
- JPEG compression (even slight)
- Image resizing
- Color space conversions
- Gamma adjustments
- Format conversions
**Solution:**
1. Export the watermarked image as PNG
2. Open that PNG file directly in chat (don't screenshot)
### Issue E: TrustMark Application Issues
**Problem:** Watermark wasn't properly embedded.
**Verify watermark was applied:**
```bash
# Test with the Python TrustMark library directly
python3 << 'EOF'
from trustmark import TrustMark
from PIL import Image
tm = TrustMark(verbose=True, model_type='Q')
img = Image.open('/path/to/your/screenshot.png').convert('RGB')
wm_secret, wm_present, wm_schema = tm.decode(img)
if wm_present:
print(f"✅ Watermark found: {wm_secret}")
else:
print("❌ No watermark detected")
EOF
```
## Step 3: Enable Detailed Logging
The MCP server logs to stderr. To see detailed TrustMark detection logs:
1. Check your MCP server logs in Cursor
2. Look for lines containing `[trustmark-service]`
3. Check for error messages
**Common log messages:**
- `TrustMark watermark detected` - Success!
- `No TrustMark watermark detected in image` - Not found (but no error)
- `TrustMark detection failed` - Error occurred
- `Python or TrustMark decoder not found` - Installation issue
## Step 4: Test with Known Watermarked Image
Create a test watermarked image:
```python
from trustmark import TrustMark
from PIL import Image
# Create test image
img = Image.new('RGB', (512, 512), color='white')
# Apply watermark
tm = TrustMark(verbose=True, model_type='Q')
watermarked = tm.encode(img, 'https://example.com/manifest.json')
# Save
watermarked.save('/tmp/test-watermarked.png')
print("Test image saved to /tmp/test-watermarked.png")
```
Then test detection:
```bash
node test-trustmark.js /tmp/test-watermarked.png
```
## Step 5: Understanding TrustMark Limitations
### What TrustMark CAN survive:
- ✅ Screenshots (in theory, if lossless)
- ✅ PNG format conversions
- ✅ Social media sharing (some platforms)
- ✅ Slight cropping
### What TrustMark CANNOT survive:
- ❌ Heavy JPEG compression
- ❌ Significant resizing (especially downscaling)
- ❌ Filters or image effects
- ❌ Color adjustments beyond minor changes
- ❌ Conversion through some image editors
## Step 6: Verify MCP Server is Using Latest Code
```bash
cd /Users/nhurwitz/mcp-content-credentials
npm run build
# Restart MCP server in Cursor
# Command Palette > MCP: Restart Server > content-credentials
```
## Need More Help?
If none of these solutions work, provide:
1. Output of `node test-trustmark.js <your-image>`
2. How the watermark was applied
3. How the screenshot was taken
4. Image format and size
5. Any error messages from MCP server logs
## Technical Details
**Detection Flow:**
1. MCP server receives file path
2. Checks for embedded C2PA manifest first (via c2patool)
3. If no manifest found, checks for TrustMark watermark
4. Calls Python script: `scripts/trustmark-decode.py`
5. Python script uses TrustMark library to decode
6. Returns JSON result back to MCP server
**Script Location:** `/Users/nhurwitz/mcp-content-credentials/scripts/trustmark-decode.py`
**Python Module:** `trustmark` (must be installed via pip)
**Model Type:** 'Q' (default, can be changed to 'P')
