AI Station Arena — Concurrent Multi-Model & Agentic MCP Workbench

⚙️

Configure System Instructions, Use-Case Presets, Agentic Tool Loop, and Custom Schemas

Station Title

Task Mode

Use-Case Preset

System Prompt / System Instructions Sent on every turn

Temperature

Max Tokens

Top-P

Reasoning Effort

Max Tool Loops

Auto-Execute Tool Calls (Agentic Loop)

Automatically run MCP or Sandbox tools and feed results back to the model until final answer. If unchecked, pauses for Manual Step/Inspect.

Enable Built-in Browser Sandbox Tools

Exposes run_javascript_sandbox, fetch_http_json, and generate_svg_diagram for immediate multi-turn tool testing.

Force Strict JSON Object Output (response_format: json_object)

Image Size:

Custom Base / Endpoint URL Override (Optional)

Leave empty for official provider endpoints. Use to point at local Ollama (http://localhost:11434/v1/chat/completions), vLLM, LiteLLM, or custom proxies.

Station Custom Tool Schemas (JSON Array)

Saved directly to station state & localStorage

Model Context Protocol (MCP) Client Inspector

Supports Streamable HTTP (2025-03-26), Legacy HTTP+SSE (2024-11-05), and WebSocket JSON-RPC 2.0

Session: None

Discovered MCP Tools & Built-in Sandbox Tools

Interactive Tool Call Tester

Tool Name

Arguments (JSON)

Result will appear here...

Live JSON-RPC 2.0 Wire Frame Log (TX / RX)

LIVE HTML / SVG SANDBOX PREVIEW

Station Configuration & Use-Case

Model Context Protocol (MCP) Client Inspector

Discovered MCP Tools & Built-in Sandbox Tools

Interactive Tool Call Tester

Live JSON-RPC 2.0 Wire Frame Log (TX / RX)