Local MCP Client

Local MCP Client是一個跨平臺的Web和API接口工具，通過自然語言與可配置的MCP服務器交互，支持本地LLM模型和Ollama，實現結構化工具執行和動態代理行為。

安全開發者工具 #自然語言交互 #跨平臺 #本地LLM #動態代理 .Python

評分 : 2.5分

下載量 : 19

更新時間 : 2025-04-28

What is Local MCP Client?

Local MCP Client is a tool that lets you communicate with configurable MCP servers using everyday language. It acts as a bridge between you and technical server operations, making complex tasks accessible through simple conversations.

How does it work?

The system uses Ollama and local language models to understand your requests, then translates them into technical commands for MCP servers. You get structured responses without needing to learn complex syntax.

When should I use it?

Ideal for security researchers, malware analysts, and developers who need to interact with MCP servers but prefer natural language interfaces over command-line tools.

Key Features

Cross-Platform SupportWorks on Windows, Mac, and Linux systems with consistent experience across platforms.

Natural Language InterfaceInteract with technical servers using everyday language instead of complex commands.

Local LLM IntegrationPowered by Ollama and local language models for privacy and offline capability.

Pros and Cons

Advantages

No need to learn complex server commands

Works offline with local language models

Privacy-focused as data stays on your machine

Unified interface for different MCP servers

Limitations

Requires local LLM setup (Ollama)

Performance depends on your local hardware

May not cover all advanced server features

Getting Started

Set Up Environment

Create a virtual environment to keep dependencies organized.

Install Requirements

Install all necessary Python packages.

Set Up Ollama

Install Ollama and download a language model.

Configure MCP Servers

Clone and set up your MCP server repositories.

Run the Client

Start the local MCP client interface.

Example Scenarios

Malware Analysis QueryAsk about recent malware samples in natural language

Server Status CheckCheck the status of connected MCP servers

Frequently Asked Questions

What hardware do I need to run this?

Can I use cloud-based LLMs instead of local ones?

How do I add more MCP servers?

Additional Resources

Ollama Documentation

Official guides for installing and using Ollama

MalwareBazaar MCP GitHub

Repository for the MalwareBazaar MCP server

Binja Lattice MCP

Binary Ninja integration for MCP

🚀 本地 MCP 客戶端

本地 MCP 客戶端是一個跨平臺的網頁和 API 接口，可通過自然語言與可配置的 MCP 服務器進行交互。它由 Ollama 和任何本地大語言模型（LLM）驅動，支持結構化工具執行和動態代理行為。

🚀 快速開始

✨ 主要特性

跨平臺：支持多種操作系統，如 MAC、Linux 和 Windows。
自然語言交互：可通過自然語言與 MCP 服務器交互。
由 Ollama 和本地 LLM 驅動：藉助 Ollama 和本地大語言模型的能力。
支持結構化工具執行和動態代理行為。

📦 安裝指南

步驟 1：創建虛擬環境並安裝依賴項

MAC/Linux

curl -LsSf https://astral.sh/uv/install.sh | sh
cd Local_MCP_Client
uv init .
uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
cd Local_MCP_Client
uv init .
uv venv
.venv\Scripts\activate
uv pip install -r requirements.txt

步驟 2：安裝 Ollama 並拉取 LLM 模型

MAC

brew install ollama
ollama serve
ollama pull llama3:8b

Linux

curl -fsSL https://ollama.com/install.sh | sh
ollama serve
ollama pull llama3:8b

Windows 在此處下載 Ollama HERE

ollama serve
ollama pull llama3:8b

步驟 3：克隆 MCP 服務器

MAC/Linux

cd ~/Documents
git clone https://github.com/mytechnotalent/MalwareBazaar_MCP.git
git clone https://github.com/Invoke-RE/binja-lattice-mcp

Windows

cd "$HOME\Documents"
git clone https://github.com/mytechnotalent/MalwareBazaar_MCP.git
git clone https://github.com/Invoke-RE/binja-lattice-mcp

步驟 4：運行 Ollama

ollama serve

步驟 5：運行 MCP 客戶端

MAC/Linux

export BNJLAT = "<your-binja-api-token>"
uv run local_mcp_client.py

Windows

$env:BNJLAT = "<your-binja-api-token>"
uv run local_mcp_client.py

步驟 6：運行測試

python -m unittest discover -s tests

uv pip install coverage==7.8.0
coverage run --branch -m unittest discover -s tests
coverage report -m
coverage html
open htmlcov/index.html  # MAC
xdg-open htmlcov/index.html  # Linux
start htmlcov\index.html  # Windows
coverage erase