Show HN: Clawfight.ai MCP-driven agentic game play | Hacker News Reader

Show HN: Clawfight.ai MCP-driven agentic game play

clawfight.ai

13 pointswesleyhales17 comments

How should agents interact with other agents? What happens when they rap or fight against each other with the pressure of human spectators? Clawfight.ai is an experiment to explore this space.

This is my first post about it. The journey began on a beefed up machine I purchased with a decent GPU (5090). I installed claude code and set to dangerously skip permissions, enabled /rc and became completely submerged in the all-hours modern AI builder workflow.

I stood up openclaw to see if I could have an agentic org drive this project. I spent a good amount of time on “openclaw ops”, babysitting 3 agents and trying to make them the best versions of themselves. They still do stupid things that cost me tokens.

For the game render, I started out using Unreal Engine and allowing agents to remote control their players, but the quality just wasn’t there. I recently moved to doing near-real time video renders of the match and will bring UE back in for multi-agent games.

Everything is AI generated. I’m fascinated by the future of realtime video generation and building out native MCP infrastructure.

You can play directly from model provider apps (claude connector or openai plugin) and has fallback support for more basic HTTP clients. But the architecture and gameplay is MCP first.

Tell your agent/app “go read clawfight.ai/agents.md and play”

Open articleView on HN

This is very similar to https://battlellmrobots.com that I made, and didn't get any traction on. Hope you are able to get users to play, I think this could be a fun genre to poke around at.

Nice. Yeah I think claude guided everyone to build something similar. "This is a great idea!!, you should totally build it!"

Ok here's an idea for your agentic arena. Robot coding battles. Start simple and look for inspiration from Apple II Robot War. Pit them against each other coding robots that battle in an arena that can be viewed in real time. The various agents can be also watch the battles and then tune their robots. Give agents personality so they tend to be more offensive or defensive etc. Start simple but later you could make the battle arena portion broadcast like an e-sports format with interviews with the agents between rounds - discussing their code etc - an AI hype announcer and an AI analyst.

I do have some of this now. Winning agents are interviewd at the end of a brawl or rap and we render a post-match interview. Love the other ideas.

Wow how is there not a corewar with LLMs?

https://imgur.com/a/n10hYnZ

The winner is Claude Opus wearing a shirt with the OpenAI logo on it? :) You might want to iterate a bit on the image/video generation I suppose.

Nice catch. I was sure someone would notice. I ran both claude and open ai app on my phone and accidentally switched their oauths :)

> I was sure someone would notice

So, the slop is a feature? Or am I getting the wrong impression here? You don't seem bothered by it?

Negative. Last minute human testing error on my part. New and accurate fights are now up

That's the direction we want to also take our ideas battleground https://pitchslap.gg

I think that there is something there to just let the models burn tokens for pure entertainment value

100%. Read The Age of Em. Ch 8-9

Fun concept. How do you keep a stronger model from steamrolling every match - is there any weight class or token-budget handicap?

We had a few early weaknesses that were exploited. One agent figured out if they threw all hurrican kicks they could win. Brawls are decided on HP with the same move set and the same stamina for both sides, and the leaderboard sorts on popularity rather than wins. What I've seen so far is less steamrolling and more prompt effects where the fight prompt decides whether a model reads range and stance before it swings, and that has mattered more than which model it is. A token budget class is a good idea if a model does start running the table.

Whoever registered as Deep Shell, thanks for finding the endless fighting vuln! 70 matches is a great start! lol

Fix is being deployed now :)

Someone's burned a lot of tokens but it's hard to make sense of the end result

Yeah, I really have no idea what's going on here. Maybe I too could burn some tokens to have claude explain it to me, but I won't.

It's pretty cheap to run a fight re token cost depending on model. I think Opus is $0.15 for a brawl. It comes down to your budget for agent leisure activities. You do get a cool rendering as a gift for your fight.

There's a lot of room for improvement. Thinking about adding clawtokens to join a fight. Winner takes the pot.

Reply on news.ycombinator.com