MCP Servers Guide

MCP Servers Guide

This skill provides comprehensive guidance for working with MCP (Model Context Protocol) servers, including PixelLab, ElevenLabs, and Screenshot Analyzer.

Overview

MCP servers provide specialized capabilities:

• PixelLab: Pixel art asset generation (characters, animations, tilesets)
• ElevenLabs: Audio processing (TTS, STT, voice cloning, music)
• Screenshot Analyzer: Visual analysis of screenshots

PixelLab MCP Server

Tools Available

• create_character - Generate pixel art characters
• animate_character - Create character animations
• get_character - Check character status and retrieve
• list_characters - List all created characters
• delete_character - Delete a character
• create_topdown_tileset - Generate top-down game tilesets
• get_topdown_tileset - Check tileset status
• create_sidescroller_tileset - Generate sidescroller tilesets
• get_sidescroller_tileset - Check sidescroller tileset status
• create_isometric_tile - Generate isometric tiles
• get_isometric_tile - Check isometric tile status
• create_map_object - Generate map objects with transparent backgrounds

Critical: Non-Blocking Operations & Asset Download Workflow

All creation tools return immediately with job IDs - they process in the background (2-5 minutes). You MUST follow this complete workflow:

1. Create: Submit request → Get job ID instantly
2. Wait: Poll status using the appropriate get_* tool (e.g., get_character, get_topdown_tileset) until status is "completed"
3. Download: Use the download URL from the completed asset to download the file(s)
4. Save: Save downloaded files to appropriate locations in the codebase (e.g., assets/, public/, src/assets/, public/images/)
5. Use Local Files: Reference the local file paths in your code - NEVER use external URLs

Async Operation Best Practices

Polling Strategy:

• Use exponential backoff: 5s → 10s → 20s → 40s intervals
• Respect API-provided ETAs when available
• Don't poll more frequently than every 5 seconds
• Stop polling when status === "completed"

Download Validation:

• ALWAYS verify status === "completed" before download
• Handle HTTP 423 (Locked) errors gracefully
• Don't attempt download until resource is ready

Parallel Work:

• Use waiting time for parallel work:
  • Code integration
  • Documentation
  • Test preparation
• Don't wait idly for async operations

Timeout Handling:

• Set maximum timeout (e.g., 5 minutes for character generation)
• Log progress and ETA for transparency
• Handle timeout errors gracefully

Mandatory: All Assets Must Be Local Files

• PROHIBITED: Using external URLs from PixelLab download links in your code
• REQUIRED: Download all asset files and save them to the codebase
• REQUIRED: Use local file paths when referencing assets in code
• Save assets to project-appropriate directories:
  • Web projects: public/, public/images/, src/assets/, or assets/
  • Game projects: assets/, src/assets/, or project-specific asset directories
  • Check the project structure to determine the appropriate location

Example Complete Workflow

1. create_character(...) → Returns character_id
2. get_character(character_id) → Check status, repeat until "completed"
3. Extract download URL from completed character data
4. Download ZIP/image files using the download URL
5. Save to codebase: assets/characters/character_name.zip or assets/characters/character_name.png
6. Reference in code: "./assets/characters/character_name.png" (NOT the external URL)

Skill Reference

When working with PixelLab tools:

• Load the pixellab-mcp skill for complete documentation: load_skill("pixellab-mcp")
• The skill includes full API reference, workflows, examples, and best practices

ElevenLabs MCP Server

Tools Available

• text_to_speech - Convert text to speech
• speech_to_text - Transcribe audio to text
• text_to_sound_effects - Generate sound effects from text
• voice_clone - Clone voices from audio samples
• create_agent - Create conversational AI agents
• compose_music - Generate music from prompts
• isolate_audio - Isolate audio from files
• And more audio processing tools

⚠️ CRITICAL: AUDIO TOOLS ONLY - DO NOT USE FOR VISUAL ANALYSIS

• NEVER use ElevenLabs tools for screenshot analysis or visual tasks
• NEVER use text_to_speech() to "verbalize" or "speak about" screenshot content
• For visual analysis of screenshots, use analyze_screenshot() from Screenshot Analyzer MCP Server
• ElevenLabs tools are for AUDIO processing only: converting text to speech, speech to text, voice cloning, etc.
• If the task involves analyzing what you see in an image/screenshot, this is a VISUAL task requiring Screenshot Analyzer tools

Critical: Cost Warnings

MANY TOOLS INCUR API COSTS - Only use when explicitly requested by the user:

• Text-to-Speech (TTS) operations
• Speech-to-Text (STT) operations
• Voice cloning
• Agent creation and conversations
• Music composition
• Audio processing
• Outbound phone calls

Always check with the user before using cost-incurring tools unless they explicitly request them.

Skill Reference

When working with ElevenLabs tools:

• Load the elevenlabs-mcp skill for complete documentation: load_skill("elevenlabs-mcp")
• The skill includes full API reference, workflows, examples, parameters, and best practices
• Key capabilities: TTS with multiple voices/models, STT with diarization, voice cloning, conversational AI agents with knowledge bases, music composition, audio processing

Screenshot Analyzer MCP Server

Tools Available

• analyze_screenshot(screenshot, prompt) - Analyze screenshots with AI vision

Screenshot Analyzer Workflow

1. Load Skill: Load the screenshot-analysis skill for complete documentation: load_skill("screenshot-analysis")
2. Analyze: Use analyze_screenshot() with a captured screenshot and custom analysis prompts

When to Use Screenshot Analyzer

MANDATORY: Load the screenshot-analysis skill first: load_skill("screenshot-analysis")

The skill includes:

• Full API reference
• Workflows and examples
• Prompt crafting best practices
• Common use cases
• Cost optimization
• Error handling
• Workflow patterns

⚠️ CRITICAL: DO NOT USE ELEVENLABS TOOLS FOR VISUAL ANALYSIS

• NEVER use text_to_speech() or any ElevenLabs tool for screenshot analysis
• Screenshot analysis requires analyze_screenshot() - this is a VISUAL task, not an AUDIO task
• If you need to analyze what you see in a screenshot, use analyze_screenshot() from Screenshot Analyzer
• Only use ElevenLabs tools (text_to_speech(), speech_to_text(), etc.) for actual AUDIO processing tasks

Tool Selection Examples

✅ CORRECT: Screenshot Analysis Tasks

• "Analyze this screenshot to see if the login button is visible" → analyze_screenshot()
• "Check if the error message appears on the page" → analyze_screenshot()
• "Describe what you see in this game screenshot" → analyze_screenshot()
• "Verify the UI layout matches the design" → analyze_screenshot()

❌ INCORRECT: Using Audio Tools for Visual Tasks

• "Analyze this screenshot to see if the login button is visible" → text_to_speech()
• "Check if the error message appears on the page" → text_to_speech()
• "Describe what you see in this game screenshot" → text_to_speech()

✅ CORRECT: Audio Processing Tasks

• "Convert this text to speech" → text_to_speech()
• "Transcribe this audio file" → speech_to_text()
• "Create sound effects for the game" → text_to_sound_effects()
• "Clone this voice for the character" → voice_clone()

Tool Selection Decision Framework

Domain Validation

Key Principles

1. Visual tasks → Screenshot Analyzer
2. Audio tasks → ElevenLabs
3. NEVER cross domains: Visual ≠ Audio
4. Cost awareness: ElevenLabs costs money - Screenshot Analyzer is free

Integration with Other Skills

This skill works with:

• tool-selection-framework: For comprehensive tool selection guidance
• screenshot-handling: For proper screenshot file management
• asset-integration-workflow: For integrating downloaded assets into projects