MCP server, free during the beta

Every AI at once, inside your agent

Add one MCP server to Claude Code, Codex, Cursor, Gemini CLI or any MCP client. Your agent can then ask up to 6 models the same question at once, GPT, Claude, Gemini, Grok, DeepSeek, Llama and more, and get every answer back side by side with its time. It is the MultiScreenAI workspace, for agents.

Server URL (Streamable HTTP)

https://www.multiscreenai.com/api/mcp/ai

Set it up

Send a key for each provider you want to ask, as an X-Key-<provider> header. Add as many as you like; free keys work too (get free keys).

Run once in a terminal

claude mcp add --transport http multiscreenai https://www.multiscreenai.com/api/mcp/ai \
  --header "X-Key-Openai: $OPENAI_API_KEY" \
  --header "X-Key-Gemini: $GEMINI_API_KEY" \
  --header "X-Key-Groq: $GROQ_API_KEY"

~/.codex/config.toml (reads the keys from your environment)

[mcp_servers.multiscreenai]
url = "https://www.multiscreenai.com/api/mcp/ai"
env_http_headers = { "X-Key-Openai" = "OPENAI_API_KEY", "X-Key-Anthropic" = "ANTHROPIC_API_KEY", "X-Key-Gemini" = "GEMINI_API_KEY" }

.cursor/mcp.json

{
  "mcpServers": {
    "multiscreenai": {
      "url": "https://www.multiscreenai.com/api/mcp/ai",
      "headers": { "X-Key-Anthropic": "sk-ant-...", "X-Key-Gemini": "AIza..." }
    }
  }
}

~/.gemini/settings.json

{
  "mcpServers": {
    "multiscreenai": {
      "httpUrl": "https://www.multiscreenai.com/api/mcp/ai",
      "headers": { "X-Key-Openai": "sk-...", "X-Key-Anthropic": "sk-ant-..." }
    }
  }
}

Any other MCP client: point it at the URL above and add the same headers.

The tools

ask_all

One prompt to several models in parallel. Name them as provider:model (for example anthropic:claude-opus-5-5) or just provider; leave them out to ask every provider you sent a key for. Set blind for test mode: answers come back as A, B, C in a shuffled order, with the key kept separate.

list_models

Every provider, its model ids, the header for its key and whether your client sent it. Free, needs no key.

Try asking your agent: “Ask GPT, Claude and Gemini how they would fix this bug, then merge the best ideas”, or “Get a blind second opinion on this plan from three models”.

Key headers

X-Key-Openai OpenAI
X-Key-Anthropic Anthropic
X-Key-Gemini Google
X-Key-Xai xAI
X-Key-Deepseek DeepSeek
X-Key-Mistral Mistral
X-Key-Groq Groq
X-Key-Openrouter OpenRouter
X-Key-Perplexity Perplexity
X-Key-Together Together AI
X-Key-Fireworks Fireworks
X-Key-Cerebras Cerebras
X-Key-Moonshot Moonshot
X-Key-Qwen Qwen
X-Key-Meta Meta
X-Key-Nvidia NVIDIA
X-Key-Cohere Cohere
X-Key-Huggingface Hugging Face
X-Key-Sambanova SambaNova
X-Key-Siliconflow SiliconFlow
X-Key-Zai Z.ai
X-Key-Ollama Ollama Cloud
X-Key-Vercel Vercel AI Gateway
X-Key-Byteplus BytePlus ModelArk

Your keys stay yours

Keys live in your own MCP client config. Each request passes them straight to that provider and nothing is stored or logged.
You pay each AI company directly at its own price, or use free keys. MultiScreenAI adds no markup.
Free during the beta, up to 300 ask_all calls a day per address.

Looking for the BNB Chain safety checks for agents? That is a separate server: the Safety Pack.