The Problem: Most MCP implementations require local installation and configuration. If you want to test someone else's MCP server or explore what's possible, you need to clone repos, install dependencies, configure auth, etc. It's friction that kills exploration.
What I built: A web-based playground where you can:
Connect to any remote MCP server via URL AI Chat with Dual Provider Support Claude Integration: Full support for Anthropic's Claude models Gemini Integration: Complete support for Google's Gemini models Custom System Prompts: Define unique personalities and behaviors. Gemini TTS support – hear responses read aloud with Google's text-to-speech
Production-Ready Authentication:
Open Servers: Public, no-authentication endpoints Bearer Token Auth: Secure API key/bearer token authentication OAuth 2.0 Flow: Full OAuth support for enterprise MCP servers
Capability Discovery: Instant visibility into Tools, Prompts, and Resources per server
Tool Restriction: Granular control over which tools AI can access All Tools Mode: AI can use any available tool Selected Tools Mode: Restrict AI to specific tools only No Tools Mode: Pure text-based AI responses
Still in beta – there are rough edges and I'm actively iterating based on feedback.
The playground is live at : https://mcpsplayground.com/chat