geepers-mcp

MCP Orchestration Skill (formerly Dream Cascade)

The MCP Orchestration skill provides advanced multi-agent coordination patterns for Claude. It allows you to launch massive, hierarchical research swarms or task-specific agent teams to solve complex problems that require multiple perspectives or large-scale data processing.

Architecture: Dream Cascade

The primary pattern is the Dream Cascade, a 3-tier hierarchical swarm:

1. Tier 1: Belters (Workers)
   • Parallel agents (1-30+) that execute specific subtasks.
   • Each Belter has a unique specialization (Research, Analysis, Technical, etc.).
2. Tier 2: Drummers (Synthesizers)
   • Aggregation agents that synthesize every 5 Belter responses.
   • They filter noise and highlight key findings from the worker tier.
3. Tier 3: Camina (Executive)
   • Final synthesis agent that provides a high-level strategic report based on Drummer syntheses.

Features

• Hierarchical Synthesis: Automatic aggregation of insights from large swarms.
• Multi-Provider Support: Switch between Anthropic, OpenAI, Gemini, etc.
• Cost Optimization: Uses ProviderFactory to select models based on tier and task complexity.
• Structured Reporting: Generates Markdown, PDF, and DOCX reports via the reporting library.
• Parallel Execution: Handled with asyncio and semaphore-based rate limiting.

Setup

Environment Variables

Ensure the following API keys are set:

• ANTHROPIC_API_KEY
• OPENAI_API_KEY
• GEMINI_API_KEY
• (Other providers defined in src/core/llm/factory.py)

Claude Desktop Configuration

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "orchestration": {
      "command": "python3",
      "args": ["/home/coolhand/geepers/skills/source/mcp-orchestration/scripts/server.py"]
    }
  }
}

Available Tools

dream_orchestrate_research

Starts a hierarchical 3-tier research workflow.

Parameters:

• task (string, required): The research topic or question.
• num_agents (integer): Total number of Belters (default: 5).
• enable_drummer (boolean): Whether to enable synthesis tier (default: true).
• enable_camina (boolean): Whether to enable executive tier (default: false).
• provider_name (string): LLM provider to use (default: "anthropic").

CLI Usage

You can also run the orchestrator directly from the command line:

# Basic run with 5 belters
python3 scripts/orchestrator.py "What is the future of AGIs?"

# Advanced run with 15 belters, 3 drummers, and 1 camina
python3 scripts/orchestrator.py "Analyze the global semiconductor supply chain" --belters 15 --drummers 3 --caminas 1 --pdf

Internal Structure

• scripts/: Executable entry points (CLI and Server).
• src/core/orchestration/: Core logic and patterns.
• src/core/llm/: Unified provider factory.
• src/core/reporting/: Multi-format report generation.
• reference/: Examples and templates.