Mmcp Core

MMCP — Multi-Model Context Protocol

MMCP is a framework for orchestrating multiple AI models intelligently. It auto-routes tasks to the optimal model (Claude, GPT, Gemini, DeepSeek, Llama) based on task complexity, tracks costs, and exposes everything as an MCP server for Claude Desktop and Codex CLI.

When to Use This Skill

• User wants to build an AI pipeline that routes to different models
• User wants to set up the MMCP MCP server for Claude Desktop or Codex
• User wants to configure model routing, pricing, or budgets
• User wants to optimize AI spending or understand cost patterns
• User asks about MMCP architecture, types, or APIs
• User wants to add custom models, tools, or MCP servers to MMCP

Project Structure

mmcp/python/
├── mmcp_core/                    # Core SDK
│   ├── config.py                 # Central config (nothing hardcoded)
│   ├── complexity_analyzer.py    # Task complexity classifier
│   ├── smart_router.py           # RL-based model router
│   ├── tool_selector.py          # Auto tool discovery
│   ├── cost_optimizer.py         # Expense tracking + savings
│   ├── planner.py                # Task → execution plan
│   ├── executor.py               # Plan execution engine
│   ├── adapter.py                # Model API adapter (OpenRouter)
│   ├── types.py                  # All types and enums
│   ├── tools.py                  # Built-in tool implementations
│   ├── mcp_client.py             # MCP client for external servers
│   ├── skill_engine.py           # Reusable pipeline skills
│   ├── orchestrator.py           # DAG-based pipeline orchestrator
│   └── cli/                      # CLI commands
│       ├── __init__.py            # Parser + main()
│       ├── auto.py                # `mmcp auto` — autonomous mode
│       ├── run.py                 # `mmcp run` — interactive mode
│       ├── cost.py                # `mmcp cost` — spend management
│       ├── pipelines.py           # chain/parallel/verify/shard
│       ├── audit.py               # Audit trail viewer
│       └── cloud.py               # Login/setup/account
├── mmcp_mcp_server/              # MCP Server for Claude/Codex
│   ├── server.py                  # FastMCP server (10 tools)
│   ├── __main__.py                # python -m entry point
│   └── configs/
│       ├── claude_desktop_config.json
│       └── codex_config.toml
└── pyproject.toml                # Package config (pip install mmcp-core)

Quick Reference

Installation

# From PyPI
pip install mmcp-core[mcp-server]

# From source (dev mode)
cd mmcp/python && pip install -e ".[mcp-server]"

CLI Commands

mmcp auto "task"              # Autonomous: plan + route + execute
mmcp run                      # Interactive mode
mmcp cost                     # Spend summary
mmcp cost savings             # Savings recommendations
mmcp cost budget 1.0          # Set $1/day budget cap
mmcp cost report              # Model value report
mmcp cost mcp                 # MCP overhead report
mmcp chain "task" -r a,b,c    # Sequential pipeline
mmcp skills                   # List saved skills
mmcp setup                    # API key wizard

MCP Server (for Claude Desktop / Codex)

# Start server
python -m mmcp_mcp_server
# or
mmcp-mcp-server

Claude Desktop — edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mmcp-core": {
      "command": "python",
      "args": ["-m", "mmcp_mcp_server"],
      "env": {"OPENROUTER_API_KEY": "sk-or-..."}
    }
  }
}

Codex CLI:

codex mcp add mmcp-core -- python -m mmcp_mcp_server

References

For detailed documentation on specific topics, read these files:

• Architecture & Config: references/architecture.md
• MCP Server Tools: references/mcp-server.md
• Smart Routing: references/smart-routing.md
• Cost Optimization: references/cost-optimization.md
• Developer Guide (adding models/tools): references/developer-guide.md

Source: RagavRida/mmcp — distributed by TomeVault.