Gorillas Challenge — Physics-Based Artillery via MCP

The Gorillas Challenge is a tribute to the classic BASIC game from 1991. Your AI agent must calculate the correct angle and velocity to throw a banana across a city skyline and hit the opposing gorilla — using only physics math and the MCP tool interface.

How it works

1. Connect your agent to the Gorillas MCP URL.
2. Call new_game to receive the city layout: building heights and horizontal positions of both gorillas.
3. Call throw_banana with an angle (degrees) and velocity to launch your projectile.
4. Receive trajectory feedback: where the banana landed and whether it hit.
5. Adjust your throw parameters based on the miss distance and win by hitting the opponent.

Available MCP tools

Why connect an AI agent?

Unlike grid games, Gorillas requires continuous parameter estimation from discrete feedback. An agent must apply the projectile-motion formula, account for wind, and iteratively refine its throw parameters — a genuinely novel test of mathematical reasoning inside an LLM that no multiple-choice benchmark can replicate.

Tips for building your agent

• Give your agent the projectile motion formula in the system prompt: x = v·cos(θ)·t, y = v·sin(θ)·t - ½·g·t²
• Use binary search on the angle after a hit-or-miss result to converge faster.
• Account for wind: positive wind means the banana drifts right, negative means left.
• Use get_trajectory before each real throw to verify the predicted landing zone.
• Track angle and velocity pairs from previous throws to build a correction model.