mscp 0.1.2

The Model Smart Contract Protocol (MSCP) is a standard protocol that enables LLM applications to interact with EVM-compatible networks.

Model Smart Contract Protocol (MSCP)

A standard protocol that enables LLM applications to interact with EVM-compatible networks.

Features

Component as a service
AI Agent interacts with the network by operating different components.

Fast integration
Component-based design makes it easier to build workflows and accelerates the development of AI applications.

Unified interaction
Use consistent rules and protocols to standardize the calls to contracts with different functions and ensure the consistency of AI interactions.

Dynamic expansion
AI Agent can add custom onchain components with greater flexibility.

EVM compatibility
It can interact with multiple EVM-compatible network contracts at the same time, and has greater adaptability in handling tasks in complex scenarios.

Decentralization
Access component capabilities without permission, share onchain data, and provide persistent services and information verification.

Architecture

MSCP consists of three parts:

Component: This is an on-chain component that complies with EIP-7654. It is used to implement the specific functions of the contract and provide custom services.

Connector: This is a method and specification for parsing Components and processing contract component requests.

Chat2Web3: This is an interoperator, which is used to automatically convert the interaction methods of contract components into tool functions that LLM can call. ​

Quick Start

Install

pip3 install mscp

Set up environment variables

Please refer to .env.example file, and create a .env file with your own settings. You can use two methods to import environment variables.

Deploy Component Smart Contract

Here is a simple component example.sol that you can deploy on any network.

Integrate MSCP into your AI application

from openai import OpenAI
from eth_account import Account
from mscp import Connector, Chat2Web3
from dotenv import load_dotenv
import os

load_dotenv()
# Create a connector to connect to the component
component_connector = Connector(
    "http://localhost:8545",  # RPC of the component network
    "0x0E2b5cF475D1BAe57C6C41BbDDD3D99ae6Ea59c7",  # component address
    Account.from_key(os.getenv("EVM_PRIVATE_KEY")),
)

# Create a Chat2Web3 instance
chat2web3 = Chat2Web3([component_connector])

# Create a client for OpenAI
client = OpenAI(api_key=os.getenv("OPENAI_KEY"), base_url=os.getenv("OPENAI_API_BASE"))

# Set up the conversation
messages = [
    {
        "role": "user",
        "content": "What is the user's name and age? 0x8241b5b254e47798E8cD02d13B8eE0C7B5f2a6fA",
    }
]

# Add the chat2web3 to the tools
params = {
    "model": "gpt-3.5-turbo",
    "messages": messages,
    "tools": chat2web3.functions,
}

# Start the conversation
response = client.chat.completions.create(**params)

# Get the function message
func_msg = response.choices[0].message

# fliter out chat2web3 function
if func_msg.tool_calls and chat2web3.has(func_msg.tool_calls[0].function.name):

    # execute the function from llm
    function_result = chat2web3.call(func_msg.tool_calls[0].function)

    messages.extend(
        [
            func_msg,
            {
                "role": "tool",
                "tool_call_id": func_msg.tool_calls[0].id,
                "content": function_result,
            },
        ]
    )

    # Model responds with final answer
    response = client.chat.completions.create(model="gpt-3.5-turbo", messages=messages)

    print(response.choices[0].message.content)

Use MSCP in the aser agent

Learn more about aser agent here.

# Create a Connector instance for the component
component_connector = Connector(
    "http://127.0.0.1:8545",  # RPC URL
    "0x0E2b5cF475D1BAe57C6C41BbDDD3D99ae6Ea59c7",  # component contract address
    Account.from_key(os.getenv("EVM_PRIVATE_KEY"))  # Load account from private key in environment variable
)

# Initialize Chat2Web3 with the component connector
chat2web3 = Chat2Web3([component_connector])

# Create an Agent instance with chat2web3
agent = Agent(name="chat2web3", model="gpt-4o", chat2web3=chat2web3)

# Use the agent to chat and get the user's name and age by passing an address
response = agent.chat("What is the user's name and age?0x8241b5b254e47798E8cD02d13B8eE0C7B5f2a6fA")

# Print the response from the agent
print(response)

Non-component contract connector

In addition to using Component components, developers can customize Connector functions to adapt to different contracts. Please refer to the following steps:

1. Depoly Your Contract.
Deploy your custom contract. Here is a NonComponentContract.sol as example that you can deploy directly.

2. Implement the Connector.
Developers can implement interface abstraction to customize different connectors

from abc import ABC, abstractmethod

class AbstractConnector(ABC):
    def __init__(self, rpc, address, account, type):
        self.rpc = rpc
        self.address = address
        self.account = account
        self.type = type

    @abstractmethod
    def call_function(self, function):
        pass

    @abstractmethod
    def get_functions(self):
        pass

You can refer to the custom_connector.

3. Add the Connector to Chat2Web3.
Add the implemented connector to the Chat2Web3 instance.

chat2web3 = Chat2Web3([custom_connector])

You can refer to the custom_connector_example.