mmcp-core 1.0.0

One CLI that auto-routes tasks to the best AI model — Gemini, GPT, Claude, DeepSeek, Llama. No model selection needed.

MMCP Python SDK

Multiple Model Context Protocol — orchestrate AI models as a coordinated DAG.

Python port of the @mmcp/core TypeScript SDK.

Install

pip install mmcp-core

# With LangChain/LangGraph support
pip install mmcp-core[langchain]

# Development
pip install mmcp-core[dev]

Quick Start

import asyncio
from mmcp_core import MMCPOrchestrator, RoleBasedRouter, MemoryStore

async def main():
    orc = MMCPOrchestrator({
        "router": RoleBasedRouter({
            "architect": {"model_id": "claude-haiku-4-5-20251001"},
            "reviewer":  {"model_id": "claude-haiku-4-5-20251001"},
        }),
        "store": MemoryStore(),
    })

    result = await orc.run_chain(
        "Explain the observer pattern in Python.",
        ["architect", "reviewer"]
    )

    print(f"✅ {result.output}")
    print(f"🪙 Tokens: {result.total_tokens}")
    print(f"💰 Cost: ${result.total_cost_usd:.6f}")

asyncio.run(main())

DAG Operations

Operation	Signature	Description
fork	1 → N	Spawn parallel sub-contexts
merge	N → 1	Combine parent outputs
handoff	1 → 1	Pass to different model/role
shard	1 → N	Split long content
verify	1 → 2	Trust contract (challenger + synthesizer)

LangGraph Tracer — One Line

from langchain_mmcp import MMCPTracer

tracer = MMCPTracer(
    regulation_tags=["SOC2", "GDPR"],
    export_path="./mmcp-audits/"
)

# Add to ANY LangGraph or LangChain pipeline
result = app.invoke(input, config={"callbacks": [tracer]})

# Access audit trail
wire_dag = tracer.get_wire_dag()
tracer.print_summary()

Wire Format

Every execution produces a JSON WireDAG with:

• SHA-256 audit hashes per node
• Full parent DAG lineage
• Token usage and cost per node
• Regulation compliance tags
• Tamper-proof audit chain

Environment

export ANTHROPIC_API_KEY=sk-ant-...

License

MIT