// the find
AI-QL/tuui
A desktop MCP client designed as a tool unitary utility integration, accelerating AI adoption through the Model Context Protocol (MCP) and enabling cross-vendor LLM API orchestration.
An Electron + Vue3 desktop app that acts as both an MCP client and a generic LLM chat UI, letting you point it at any OpenAI-compatible or Anthropic-style endpoint and wire up local or remote MCP servers for testing. It's aimed at people building or debugging MCP servers who want a GUI inspector instead of the official CLI inspector, with zero account/cloud dependency.
Covers more of the MCP spec than most client projects bother with — tools, prompts, resources, sampling, and elicitation are all implemented, plus registry discovery and MCPB bundle support, not just the bare tool-calling path. LLM backend config is a flat, hand-editable JSON array (url/path/model/modelList) so swapping between OpenAI, Claude, Qwen, or a self-hosted endpoint is a config edit, not a rebuild. It also supports remote MCP servers over mcp-remote with OAuth, not just local stdio processes. For an Electron side project it has real CI discipline — Playwright UI tests and enforced ESLint/Prettier rules running in GitHub Actions.
LLM API keys and all config live in plaintext localStorage with no OS keychain integration, which is a real problem for a desktop app whose whole job is holding your API keys. The README is upfront that large parts of the codebase were AI-converted from a prototype rather than hand-written, and the strict linting setup is a reactive guardrail for that, not evidence the generated code is actually sound — worth a closer look before trusting it with production credentials. macOS has an open, unresolved issue with MCP server spawn (ENOENT) that the maintainers only paper over with workarounds, so non-Windows users should expect friction. Roots (the client-side MCP capability for exposing filesystem context) isn't implemented, which rules out IDE-style vibe-coding workflows.