Dive

Dive is a desktop application that acts as a host for Model Context Protocol (MCP) servers: it connects to them over stdio or SSE and pairs them with any language model that supports function calling, then gives you a chat interface for driving both. It runs on Windows, macOS, and Linux, and ships in two builds — an Electron version and a newer Tauri version with a smaller installer.

Model support covers ChatGPT, Anthropic, Ollama, and OpenAI-compatible endpoints, with several API keys and model switching configured through model_settings.json. Individual MCP tools can be enabled or disabled for finer control, custom system prompts adjust behavior, and MCP server authentication is supported, which the project marks as unstable and possibly requiring frequent re-authorization. Managed MCP servers can also be connected from the OAP cloud service through one-click deep links or configuration files, and existing local MCP and model configurations keep working alongside it.

Local tools are preconfigured: Fetch for web requests, a file manager for reading and writing files, and Bash for command execution. An installer agent helps set up and configure further MCP servers. The interface covers more than 24 languages, keyboard shortcuts for renaming, settings, reload, and new chats, chat draft saving, chat history search, skills and slash commands, an @ keyword for searching file paths, handling of several MCP elicitation requests at once, and automatic update checks.