Unreal 5.9 MCP builds jungle FPS with two prompts: Givros shows a 4h54m workflow
By GameSlash
Givros shares two detailed prompts for Temple of the Canopy: first create a navigable jungle temple with Unreal MCP, then add FPS team deathmatch and testing.
Creator Givros shared 88 seconds of an Unreal Engine 5.9 first-person shooter called Temple of the Canopy on November 19, 2026. He says a GPT 6.2 Astra workflow with Unreal MCP, Blender, Three.js and a custom game-building skill produced it in 4 hours 54 minutes, spending about 15% of his usage allowance. Those are creator-reported figures, not a GameSlash benchmark. Crucially, he published his two detailed prompts so developers can study the separation of level building and gameplay design. Original demonstration and prompts.

Prompt one: a map, not just a decorative facade
First, Givros used a ChatGPT-generated reference picture of a tropical stone temple, pools, waterfalls, bridges, huge tree roots and jungle vegetation. His first long prompt instructs Unreal MCP to use the image as the primary visual target, repeatedly inspect its current level, align camera angles, compare screenshots and correct visual mismatches. It insists on connected areas beyond the single reference view and replacement of visible blockout primitives with finished assets. Blender is named for asset production, while the exact purpose of Three.js is not explained.

Prompt two: convert the environment into a team deathmatch
His second prompt preserves the jungle-temple map but adds a red-versus-blue FPS design: two teams with separate spawns, basic combat, hit detection, damage, death and respawn, a scoreboard, AI opponents and a match-ending condition. Example parameters in the prompt are 50 kills or 10 minutes. It also demands navigation checks, spawn safety, gun recoil, reload, bot matches and repeated playtests. These are specifications in the prompt, not independent evidence every feature passed testing.

The useful rule for AI game building
Instead of asking an agent to 'make a complete FPS' in a single vague request, specify a reference-guided environment pass, then a separate mechanical acceptance checklist and testing loop. The creator's custom skill is not published with the post, so the two prompts do not guarantee the same result on other setups. The short video also does not independently establish that every navigation route, spawn or match-ending condition works; treat this as a developer demonstration rather than a completed multiplayer review.
Primary source: Givros's original X thread and both prompts.