Dive

Dive is an open-source MCP Host Desktop Application designed to seamlessly integrate with any large language model (LLM) that supports function calling. It acts as a bridge between your LLMs and pre-configured MCP tools, enabling you to run modular MCP servers and manage multiple models and tools from a single interface. Dive supports secure MCP server authentication, multi-language prompts, and built-in local tools such as Fetch, File Manager, and Bash, making it suitable for developing and testing agents that need to fetch data, read or write files, and execute shell commands. The application also includes an MCP Server Installer Agent to assist with installing and configuring MCP servers automatically, reducing manual setup time.

To use Dive, install the MCP server via the provided setup options (either locally or via OAPHub cloud services), then connect your preferred LLM to the MCP server. You can enable or disable individual MCP tools, adjust model settings with the model_settings.json, and manage multiple elicitation requests concurrently from the UI. For cloud-enabled workflows, you can leverage OAPHub.ai to deploy and manage MCP servers without worrying about local dependencies. Dive supports both the modern Tauri architecture and the traditional Electron version, giving you flexibility based on your platform and performance needs.

In practice, you will typically sign up or sign in to OAPHub, connect a Dive MCP server, and begin issuing MCP requests from your chosen LLM, while Dive handles tool calls, context, and prompts according to your configured settings.

Prerequisites:

• A supported operating system (Windows, macOS, or Linux)
• Node.js and npm (preferably the latest LTS) or access to a prebuilt Dive release
• Optional: Rust toolchain for building the Tauri variant (cargo, rustc, etc.) if you plan to build from source
• Access to internet to download dependencies

Install steps:

1. Clone or download the Dive repository from GitHub or use the provided release artifacts.
2. Install dependencies:

# If using npm-based workflow
npm install

3. Run the development environment or build the app:

# Electron (development)
npm run dev

# Tauri (development)
cargo tauri dev

4. If you plan to run the MCP server locally with npx, ensure you have Node.js and npm installed. You can start the MCP server using the recommended npx command from the mcp_config:

npx -y openagentplatform-dive

5. For cloud-based usage via OAPHub, follow the README's MCP Setup Options to sign up and connect your Dive instance to an MCP server managed by OAPHub. This typically involves configuring the Dive client to connect to the chosen MCP server and providing any required API keys.

Recommended approach:

• On first run, allow the app to install its bundled Python/Node environments automatically if prompted.
• If you prefer local environment control, pre-install Python and Node.js in your system and use the provided setup flow to register Dive as an MCP server.

Tips and common considerations:

• If you enable authentication for MCP servers, expect occasional re-authorization due to token expiry; ensure your app can prompt for re-auth without interrupting workflows.
• When using OAPHub integration, keep your API keys secure and rotate them periodically.
• The model_settings.json file allows granular control over model and tool usage; adjust it to balance latency, throughput, and tool availability.
• If you encounter sandbox or permissions issues on Linux (especially with AppImage), you may need to run with --no-sandbox or grant executable permissions to the binary.
• The Dive MCP server supports multiple elicitation requests concurrently in the UI, so monitor resource usage to avoid bottlenecks on CPU or memory.
• If you customize or replace local tools (Fetch, File Manager, Bash), ensure proper security controls to prevent unintended access to the host system.