suno-upload（苏诺上传）

Suno Upload

Automatically upload Suno prompt files to suno.com using Chrome automation. This skill parses your prompt.md files and fills the Suno Create interface with all the necessary fields.

✅ Fully Automated - No Human Intervention Required

This skill achieves complete end-to-end automation of the Suno song creation process:

• ✅ Parses prompt.md files - Extracts title, lyrics, style, model, and all parameters
• ✅ Navigates to Suno automatically - Opens suno.com/create and switches to Custom mode
• ✅ Fills ALL form fields - Including complex React slider components
• ✅ Slider automation working - Uses coordinate-based dragging for precise control
• ✅ Submits and monitors - Clicks Create button and tracks generation
• ✅ Validated and tested - Successfully generated songs with full parameter control

No manual intervention needed - Just point it at a prompt.md file and let it work.

Prerequisites

• Chrome MCP server must be installed and active
• Suno account with available credits
• Prompt.md file in the standard Suno song creator format

When to Use

Use this skill to:

• Upload generated prompt.md files to Suno
• Automate the song creation process on suno.com
• Submit multiple prompts efficiently
• Avoid manual copy-paste of lyrics and settings

Workflow

Step 1: Locate Prompt Files

Search for prompt.md files in the current directory and subdirectories.

If multiple files found:

• Display list with:

- File path - Song title (from YAML frontmatter) - Project name (from YAML frontmatter) - Last modified date

• Use AskUserQuestion for interactive selection
• Present options clearly with full context

If single file found:

• Automatically select it
• Display which file was selected
• Show basic info (title, project)

If no files found:

• Error message: "No prompt.md files found in the current directory or subdirectories."
• Guidance: "Please navigate to a directory containing Suno prompt files, or create a prompt using the suno-song-creator skill first."
• Exit gracefully

Tools used:

• Glob - Search for **/prompt.md pattern
• Read - Extract YAML frontmatter for file listing
• AskUserQuestion - Interactive file selection

Step 2: Parse Prompt File

Read and extract data from the selected prompt.md file.

Parsing Strategy:

1. YAML Frontmatter (lines 1-12):

---
title: "Song Title Here"
project: "Project Name"
---

• Extract title: value
• Remove surrounding quotes if present
• This becomes the "Song Title" field

2. Structured Prompt (under "### Structured Prompt" heading):

Find the heading ### Structured Prompt, then extract everything between the triple backticks:

genre: "bubblegum pop, synth-pop, disco-influenced pop..."
vocal: "bright female pop vocals, playful delivery..."
instrumentation: "synth bass, disco-inspired drums..."
production: "polished modern pop, wide stereo mix..."
mood: "playful, sarcastic, witty, upbeat..."

IMPORTANT:

• Copy the entire block as-is (all 5 lines)
• Do NOT parse individual fields
• This complete block goes into Suno's "Styles" field
• Keep all quotes and formatting

3. Lyrics Section (after "## Lyrics" heading):

Find ## Lyrics, then extract everything between the triple backticks:

///*****///

[Verse 1 | playful | bright production]
Lyrics text here...

[Chorus | upbeat | full production]
Chorus lyrics here...

CRITICAL Parsing Rules:

• REMOVE the divider line: Skip ///*****/// at the start (this is just a separator in the file)
• KEEP all meta tags: [Verse 1 | playful | bright production] - Suno uses these!
• KEEP section markers: [Verse], [Chorus], [Bridge], etc.
• KEEP all blank lines: Preserve formatting exactly
• Start from first actual lyric content: After removing divider, begin with the first [Verse...] tag

Example extraction:

File contains:
///*****///

[Verse 1 | playful | bright production]
You showed up late...

Extract as:
[Verse 1 | playful | bright production]
You showed up late...

4. Model Selection (from "### Model and Parameters" section):

Look for the model specification near the top of this section:

**Model:** v5 (cleanest audio, most natural vocals - best for modern pop)

Extraction rules:

• Model: Extract value after **Model:** and before the opening parenthesis

- Expected values: "v5", "v4.5", "v4.5+", "v4" - Trim whitespace: "v5" - Default to "v5" if not specified

5. Parameters (from "### Model and Parameters" section):

Look for bullet points under this heading:

- Weirdness: 40%
- Style Influence: 60%
- Vocal Gender: Female
- Exclude Styles: Rock, Metal, Country

Extraction rules:

• Weirdness: Extract number before % (e.g., "40" from "40%")

- If range like "30-40%", use midpoint (35) - Convert to slider value: use as-is (0-100 scale)

• Style Influence: Extract number before % (e.g., "60" from "60%")

- If range like "60-70%", use midpoint (65) - Convert to slider value: use as-is (0-100 scale)

• Vocal Gender: Extract value after : and trim whitespace

- Expected values: "Female", "Male", or unspecified - Map to Suno's gender buttons

• Exclude Styles: Extract comma-separated list after :

- Trim whitespace: "Rock, Metal, Country" - Keep as comma-separated string

Graceful Handling:

• If field missing: use sensible default or skip
• If malformed YAML: try regex extraction as fallback
• If lyrics section empty: error (lyrics required for Suno)
• If parameters missing: use Suno defaults (50% for sliders)

Tools used:

• Read - Read the prompt.md file
• Regex/string parsing - Extract structured data

Validation:

• Title exists and non-empty
• Lyrics exist and non-empty
• Structured prompt exists and non-empty
• Parameters are within valid ranges (0-100 for sliders)

Step 3: Display Parsed Data

Show the user what was extracted from the file for verification:

Format:

📋 Parsed Prompt Data:

Title: "Fixer-Upper"
Project: "Pop Songs I Love"
Model: v5

Lyrics Preview (first 200 chars):
[Verse 1 | playful | bright production]
You showed up late with pizza stains
On that shirt you wore last Tuesday
Asked me if I'd do your laundry
While you played your gam...

(Note: Divider line removed for Suno upload)

Style/Genre (from structured prompt):
genre: "bubblegum pop, synth-pop, disco-influenced pop..."
[Full 5-line structured prompt shown]

Parameters:
- Weirdness: 40%
- Style Influence: 60%
- Vocal Gender: Female
- Exclude Styles: Rock, Metal, Country

Ready to upload to Suno!

Purpose:

• User can verify parsing was correct
• Catch any issues before automation starts
• Builds confidence in the process

Step 4: Initialize Chrome Session

Set up the browser automation environment.

Steps:

1. Call tabs_context_mcp to get current tab state

- Check if MCP tab group exists - Get available tab IDs

1. If no MCP tab group: call tabs_context_mcp with createIfEmpty: true
2. Create new tab: tabs_create_mcp

- Fresh tab for Suno automation - Prevents interfering with user's existing tabs

1. Navigate to suno.com: navigate(url: "https://suno.com")
2. Wait for page load: computer(action: wait, duration: 3)

Error handling:

• If navigation fails: retry once, then ask user to check connection
• If tab creation fails: inform user and exit
• If already on suno.com: ask if should create new tab or use existing

Tools used:

• tabs_context_mcp - Get tab context
• tabs_create_mcp - Create new tab
• navigate - Navigate to URL
• computer (wait) - Wait for page load

Step 5: Navigate to Create Interface

Navigate from the home page to the Create page in Custom mode.

Steps:

1. Navigate to Create page: - Use navigate tool: "https://suno.com/create" - OR use find + computer tools to click "Create" button - Wait 2 seconds for page load
2. Switch to Custom mode: - Use find tool: query="Custom tab button" - Use computer tool: left_click on the Custom button ref - Wait 1 second for mode switch
3. Verify Custom mode loaded: - Take screenshot for debugging (optional) - Use read_page to verify "Lyrics" field is present

UI Elements to find:

• Create button/link (ref varies, use find tool)
• Custom tab button (vs "Simple" mode)
• Lyrics textarea (confirms Custom mode loaded)

Error handling:

• If Create page doesn't load: retry navigation once
• If Custom button not found: take screenshot, inform user
• If timeout: ask user to check Suno status

Tools used:

• navigate - Direct URL navigation
• find - Locate UI elements by description
• computer - Click buttons
• read_page - Verify page loaded correctly

Step 6: Fill Form Fields

Fill each form field with the parsed data.

Field-by-field process:

a. Model Selector (at top of Create area)

- Use find: query="v5" or "model selector dropdown"
- Location: Top of create area, near Simple/Custom tabs
- Current default: v5
- Click to open dropdown if not already showing target model
- Select the parsed model (v5, v4.5, v4.5+, or v4)

Data: Model from Step 2 (e.g., "v5") Note: Based on UI exploration, the model selector is visible at the top Implementation:

• If model is "v5" (default): may skip (already selected)
• If model is different: click dropdown and select appropriate option
• Use find to locate the specific model option
• Use computer.left_click to select

b. Lyrics Textarea

- Use find: query="lyrics textarea"
- Use form_input: ref=<lyrics_ref>, value=<full_lyrics_text>
- Or use computer.type if form_input fails

Data: Full lyrics from Step 2 (with divider line ///*****/// removed, but all meta tags like [Verse 1 |...] preserved)

c. Styles Field (Structured Prompt)

- Use find: query="styles textbox" or "style of music"
- Use form_input: ref=<styles_ref>, value=<structured_prompt_block>

Data: Complete 5-line structured prompt block

d. Song Title (Optional)

- Use find: query="song title optional"
- Use form_input: ref=<title_ref>, value=<song_title>

Data: Title from YAML frontmatter

e. Expand Advanced Options

- Use find: query="Advanced Options"
- Use computer: left_click on Advanced Options button
- Wait 1 second for expansion

f. Exclude Styles

- Use find: query="exclude styles"
- Use form_input: ref=<exclude_ref>, value=<exclude_styles_list>

Data: Comma-separated exclusion list (e.g., "Rock, Metal, Country")

g. Vocal Gender

- Use find: query="Female" (or "Male" based on parsed data)
- Use computer: left_click on the appropriate gender button

Data: "Female" or "Male" from parameters Note: If unspecified, skip this field (use Suno default)

h. Weirdness Slider

Data: Parsed weirdness value (0-100)

Working Method (Coordinate-Based Dragging):

Sliders in Suno are custom React components (DIVs with ARIA roles), not standard HTML inputs. They require coordinate-based mouse dragging to set values.

Step-by-step implementation:

1. Find the slider element: Use find: query="Weirdness slider" This returns a ref to the slider with role="slider"
2. Calculate target position with JavaScript: const slider = document.querySelector('[role="slider"][aria-label*="Weirdness"]'); const thumb = slider.children[slider.children.length - 1]; const sliderRect = slider.getBoundingClientRect(); const thumbRect = thumb.getBoundingClientRect(); // Current thumb center position const startX = thumbRect.left + thumbRect.width / 2; const startY = thumbRect.top + thumbRect.height / 2; // Target position (e.g., 40% along slider width) const targetPercent = 40; // Use parsed value const targetX = sliderRect.left + (sliderRect.width * (targetPercent / 100)); const targetY = sliderRect.top + sliderRect.height / 2; Use javascript_tool to execute this and return coordinates
3. Drag thumb to target position: Use computer: left_click_drag - start_coordinate: [startX, startY] (from JavaScript) - coordinate: [targetX, targetY] (from JavaScript)

Why this works:

• React state updates when the mouse drag event completes
• Coordinate calculation ensures accurate positioning
• getBoundingClientRect() provides precise pixel positions
• Dragging the thumb (last child element) triggers React's onChange handler

i. Style Influence Slider

Data: Parsed style influence value (0-100)

Working Method (Same as Weirdness):

Use the identical coordinate-based dragging approach:

1. Find the slider: Use find: query="Style Influence slider"
2. Calculate coordinates with JavaScript: const slider = document.querySelector('[role="slider"][aria-label*="Style Influence"]'); const thumb = slider.children[slider.children.length - 1]; const sliderRect = slider.getBoundingClientRect(); const thumbRect = thumb.getBoundingClientRect(); const startX = thumbRect.left + thumbRect.width / 2; const startY = thumbRect.top + thumbRect.height / 2; const targetPercent = 60; // Use parsed value const targetX = sliderRect.left + (sliderRect.width * (targetPercent / 100)); const targetY = sliderRect.top + sliderRect.height / 2;
3. Drag to target: Use computer: left_click_drag from [startX, startY] to [targetX, targetY]

Important notes for both sliders:

• Do NOT use form_input (fails with "Element type 'DIV' is not a supported form input")
• Do NOT set ARIA attributes directly (doesn't update React state)
• Do NOT use keyboard events (Arrow keys don't trigger state updates)
• DO use left_click_drag with calculated coordinates
• Precision: Usually accurate within ±1%, which is acceptable

Implementation Notes:

• Fill fields sequentially, not in parallel
• Wait 0.5 seconds between field fills for UI stability
• Take screenshot after critical fields for debugging
• If field not found, log warning and continue (optional fields)
• If required field fails, error and exit

Tools used:

• find - Locate form elements
• form_input - Fill text fields
• computer (left_click) - Click buttons, interact with sliders
• computer (type) - Fallback for text input
• computer (wait) - Stability pauses

Step 7: Confirmation & Review

Before submitting, show the user what will be submitted and ask for confirmation.

Process:

1. Take screenshot of filled form: - Use computer: screenshot action - Capture the entire Create form area - User can visually verify all fields
2. Display summary to user: ✅ Suno Form Ready for Submission: Title: "Fixer-Upper" Lyrics: 140 lines (with meta tags preserved) Style: bubblegum pop, synth-pop, disco-influenced pop... Weirdness: 40% Style Influence: 60% Vocal Gender: Female Exclude Styles: Rock, Metal, Country [Screenshot of filled form displayed above]
3. Ask for confirmation: Use AskUserQuestion: - Question: "Submit this song to Suno for generation?" - Options: * "Yes, create the song" (proceed to submit) * "No, cancel" (exit without submitting) - Wait for explicit user response
4. Handle response:

- If "Yes": proceed to Step 8 - If "No": inform user and exit gracefully - Message: "Upload cancelled. The form has been filled but not submitted. You can manually review and submit in the browser if needed."

Why this matters:

• User has final control before spending credits
• Catch any parsing errors before submission
• Visual confirmation builds trust
• Prevents accidental submissions

Tools used:

• computer (screenshot) - Capture form
• AskUserQuestion - Get explicit confirmation

Step 8: Submit & Monitor

Submit the form and monitor for song URLs.

Submission:

1. Find Create button: - Use find: query="Create song button" or "Create button" - Expected ref like: ref_943 (based on exploration)
2. Click Create button: - Use computer: left_click on Create button ref - Wait 3 seconds for submission processing
3. Monitor for response: - Page may redirect or update with song generation status - Look for success indicators or error messages - Wait up to 10 seconds for initial response

URL Extraction (if generation starts):

1. Check for song URLs: - Suno typically generates 2 songs per creation - URLs may appear immediately or after processing - Format: https://suno.com/song/[song-id]
2. Extract URLs: - Use read_page to get page content - Look for song URLs in the recent creations area - OR check the workspace/library section
3. Display results to user: ✅ Song submitted to Suno! Generation started. Suno is creating your song(s). URLs (check these for your songs): - https://suno.com/song/abc123-def456-... - https://suno.com/song/xyz789-uvw012-... Note: Songs may take 1-2 minutes to generate. Visit the URLs above to listen once ready.

Error Handling:

• Submit button not found:

- Take screenshot - Error: "Could not find Create button. Suno UI may have changed." - Offer to show screenshot to user for manual completion

• Validation errors from Suno:

- Read error messages from page - Display to user: "Suno validation error: [error message]" - Suggest fixes if recognizable (e.g., "Lyrics too short")

• Network/timeout errors:

- Error: "Submission timed out. Please check Suno manually." - Provide form status: "Form was filled and Create button clicked, but response not received."

• Credits exhausted:

- If error message about credits appears - Inform user: "Suno credits exhausted. Please purchase more credits and try again."

URL Extraction Fallback:

If URLs cannot be automatically extracted:

• Inform user: "Song submitted successfully, but URLs could not be extracted automatically."
• Guidance: "Check your Suno workspace at https://suno.com/me for your newly created songs."

Tools used:

• find - Locate Create button
• computer (left_click) - Submit form
• computer (wait) - Wait for processing
• read_page - Extract URLs and errors
• computer (screenshot) - Debugging

Step 9: Completion

Wrap up the automation and provide final status.

Success case:

✅ Suno Upload Complete!

Song: "Fixer-Upper"
Status: Submitted and generating
URLs:
- [Song URL 1]
- [Song URL 2]

Songs typically generate in 1-2 minutes. Visit the URLs above to listen!

Partial success case:

✅ Suno Upload Submitted!

Song: "Fixer-Upper"
Status: Form submitted, URLs not yet available

Check your Suno workspace for the generated songs:
https://suno.com/me

Failure case:

❌ Suno Upload Failed

Error: [Specific error message]

The form was partially filled. You can:
1. Complete manually in the browser (tab is still open)
2. Try running /suno-upload again
3. Check the error message above for guidance

Error Handling Patterns

File Not Found:

• Clear message about what went wrong
• Guidance on how to fix (navigate to correct directory, create prompt first)
• No stack traces or technical jargon

Parsing Errors:

• Show which field failed to parse
• Show the problematic content if helpful
• Suggest fixing the prompt.md file
• Offer to proceed with defaults if non-critical field

Chrome Automation Failures:

• Take screenshot at failure point
• Show screenshot to user (visual debugging)
• Provide context about what was being attempted
• Offer retry option if transient failure likely

Suno UI Changes:

• If element not found after reasonable attempts
• Error: "Suno UI may have changed. This skill needs updating."
• Recommend manual upload as fallback
• Ask user to report issue

Network Issues:

• Distinguish between local network and Suno server issues
• Suggest checking internet connection
• Offer retry option
• Timeout values should be generous (10+ seconds for submission)

Usage Examples

Example 1: Single prompt file

User: "Upload this to Suno"
Skill:
- Finds single prompt.md in current directory
- Parses: "Fixer-Upper" pop song
- Displays parsed data
- Navigates to Suno, fills form
- Asks confirmation
- User: "Yes"
- Submits and returns URLs

Example 2: Multiple prompt files

User: "/suno-upload"
Skill:
- Finds 5 prompt.md files in subdirectories
- Lists all 5 with titles and projects
- Asks which to upload
- User selects #3
- Proceeds with upload workflow

Example 3: Called from suno-song-creator

User: Creates song with suno-song-creator
Skill (suno-song-creator):
- Generates prompt, saves to prompt.md
- Asks: "Upload to Suno now?"
- User: "Yes"
- Invokes suno-upload skill with file path
- suno-upload proceeds automatically

Tools Reference

Discovery & Parsing:

• Glob - Find prompt.md files
• Read - Read file contents
• AskUserQuestion - File selection, confirmation

Chrome Automation:

• tabs_context_mcp - Get tab state
• tabs_create_mcp - Create new tab
• navigate - Navigate to URLs
• computer (screenshot) - Capture screenshots
• computer (wait) - Stability pauses
• computer (left_click) - Click buttons
• computer (type) - Type text (fallback)
• find - Locate UI elements by description
• form_input - Fill form fields
• read_page - Get page structure

Success Criteria

✅ Finds and parses prompt.md files correctly ✅ Extracts all fields without data loss ✅ Preserves lyrics meta tags and formatting ✅ Navigates to Suno Create interface ✅ Fills all form fields accurately ✅ Handles sliders and special inputs ✅ Shows confirmation before submission ✅ Submits successfully and extracts URLs ✅ Handles errors gracefully with clear messages ✅ Works when called from other skills

Known Limitations

Slider Precision:

• Coordinate-based dragging achieves accuracy within ±1%
• This variance is negligible and acceptable for creative parameters
• Browser rendering and viewport size don't affect the calculation (uses getBoundingClientRect)

URL Extraction:

• Suno's generation is asynchronous
• URLs may not appear immediately
• Fallback: direct user to workspace

UI Changes:

• Suno may update their UI
• Skill may need updates when UI changes
• Reference selectors may break

Credit Requirements:

• Skill does not check credit balance beforehand
• User must have credits for submission to work
• Error appears only at submission time

Maintenance Notes

If Suno UI changes, update these sections:

1. Step 5: Navigation paths and button refs
2. Step 6: Form field selectors and names
3. Step 8: Create button selector
4. URL extraction patterns

Test with real prompt.md files regularly to ensure parsing remains accurate.

Version: 2.0.0 - Complete Automation Achieved Last Updated: 2026-01-01 Dependencies: Chrome MCP server, Suno account with credits

Key Improvements in v2.0:

• ✅ Slider automation fully working with coordinate-based dragging
• ✅ Complete end-to-end automation with NO human intervention required
• ✅ Reliable React component interaction through mouse events
• ✅ Validated with successful song generation ("Fixer-Upper" test case)