cranebeaker/coop-anything

The problem

You want to play a single-player game with friends, but no co-op mod exists, or the one that does is abandoned and half-broken. Historically this means months of reverse engineering, engine archaeology and netcode work — work most players can't do and most modders won't finish. Co-op Anything hands that work to an AI coding agent: point it at the repo and it researches, reverse engineers, builds or revives the mod, tests it in-game with bots, and packages a one-click install.

What's different

Normally you'd search Nexus or Thunderstore for a mod, or join a Discord and hope a maintainer picks up a dead project. Co-op Anything instead ships a playbook and a set of agent skills (coop-reverse-engineer, coop-networking, coop-revive) plus an MCP server, so the agent — Claude Code, Codex, Cursor and similar — has the method, not just the prompt. Its second differentiator is the testing toolbox: chaos, flight, fuzz, ownership and animcheck give you lag injection, desync bisection, invariant fuzzing and pose comparison across real game clients. That is closer to netcode QA infrastructure than to a mod list, and it is what makes "agent writes the mod" plausible rather than aspirational. There is no direct competitor doing agent-driven co-op conversion; the closest neighbours are Nucleus Co-op and PartyDeck for split-screen, and general RE MCPs like REA for inspection only.

Who it's for

Patient PC players with an AI coding agent already set up, who are willing to run experiments and iterate — especially those chasing older titles (Deus Ex, Dungeon Keeper, Fallout: New Vegas) where existing mods can be revived. It suits modders who want a structured methodology. It is not for console-only players expecting a turnkey tool, non-technical users who just want to click play, or anyone looking for anti-cheat-safe online play; the README is honest that hard binary games are the hardest path.

Cost of adoption

You need a capable coding agent, a cloned repo, Python, and usually a full reverse-engineering toolchain (Ghidra, x64dbg, REA, engine-specific SDKs and mod loaders) depending on the game. Runs are Windows-centric; Steam relay networking is the default for joining. The license is MIT, but the actual co-op output lives in per-game repos of varying quality, and the project is eight days old. Real risk: agent-generated netcode can be subtly wrong, and "tests pass" does not guarantee a fun or stable session.

Verdict

This is a brand-new project riding a sharp spike: 8 days old, roughly +135 stars per day since creation, and #18 on the new Python board, but no daily trending appearance and only a single low-traffic Hacker News thread. The idea is genuinely novel and the supporting test tooling looks substantive, yet almost nothing is proven at scale. Treat it as an experiment worth watching and possibly contributing to, not as an install-and-play solution. If you have an agent and a stubborn favourite game, the cost of trying one title is low; if you want guaranteed co-op tonight, look elsewhere for now.