Co-op Anything
Point a coding agent at a single-player game and it builds or fixes co-op so you can play with friends.
- Source repo
- cranebeaker/coop-anything
- Stars
- ★ 381
- Last updated
- 1d ago
- License
- MIT
- Primary language
- C
- FA score
- 37/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- A player who wants story co-op in Fallout: New Vegas or Deus Ex and needs an existing mod revived through Path 1
- Not a fit if
- Players who want a one-click mod and no coding agent in the loop
- Air-gapped setups that cannot fetch tools and dependencies online
- Teams targeting anti-cheat or live-service multiplayer games
- Source review
- 37/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Co-op Anything is a repository-shaped workbench for AI coding agents: AGENTS.md rules, a skill set under .agents/skills, a playbook of method guides, a DIRECTORY.md of 230+ co-op mods and gaps, and a Python MCP server at tools/mcp/coop_mcp.py. Given a game, the agent picks one of two paths — revive a stalled co-op mod, or reverse engineer the game and its engine, hook in, sync a second player plus enemies/world/story, and handle saves and level changes. Work ends with bot-driven automated QA against real game clients, Steam Spacewar relay networking so friends join without port forwarding, one-click packaging, and a pull request back to this repo. The repo hosts no game files or assets; each game's code and releases live in its own repository. It is a tool for tinkerers and contributors, not a finished mod you install blindly.
The agent follows the phase map in AGENTS.md and picks skills such as coop-research, coop-revive, coop-qa, coop-reverse-engineer, coop-build-from-zero and coop-networking. On Path 1 it locates the last working build, mines issues, forums and Discord for bug reports, audits the code, fixes it as a separate package and ships. On Path 2 it reverse engineers the game and engine, hooks into it, renders player two, syncs enemies and world state, reintegrates an engine's existing netcode (Unreal, Source, X-Ray) or adds C# networking under BepInEx/MelonLoader, and for console titles uses decomps, recompilers or emulator bridges. The Co-op Anything MCP server searches the directory, detects a game's engine from its install folder, serves the playbook and skills, and validates contributions; it loads via .mcp.json or claude mcp add and pairs with reverse-engineering MCPs like REA. Outputs include fix packages, installable releases, Steam relay configuration, bot-driven tests, and directory entries tagged with a QA level.
- A player who wants story co-op in Fallout: New Vegas or Deus Ex and needs an existing mod revived through Path 1
- A tinkerer adding netplay to a decompiled or recompiled N64/PS1 port
- Someone setting up split-screen couch co-op on one PC via Nucleus Co-op or PartyDeck
- A community contributor filing a fix, bug list or test for a game and opening a draft PR
- A tool builder using the MCP server to query the co-op directory, detect engines and validate entries
How do you install or deploy this agent?
Give the repository URL to a supported coding agent (Claude Code, Codex, Gemini CLI, Cursor, OpenCode), which then follows AGENTS.md. The MCP server needs Python:
git clone https://github.com/cranebeaker/coop-anything
python <clone>/tools/mcp/coop_mcp.pyClaude Code picks it up from the repo's .mcp.json; for other agents add it manually:
claude mcp add coop-anything -- python <clone>/tools/mcp/coop_mcp.pyFor binary reverse engineering, pair it with REA:
npx rea-agents setupThe repo does not state a Python version requirement or ship a packaged installer; you must supply those details yourself.
How do you use this agent?
Prompt your agent to read the repo and follow AGENTS.md:
Read https://github.com/cranebeaker/coop-anything and follow AGENTS.md.
I want to play <game> in co-op with my friends.Codex, Gemini CLI and Cursor MCP setup is in playbook/mcp.md; networking and the Steam Spacewar relay are in playbook/networking.md and playbook/networking-steam-spacewar.md; bot-driven testing is in playbook/automated-qa.md. Check DIRECTORY.md first to see whether a co-op mod already exists. The agent asks before installing anything or using your accounts, and returns its work as a draft pull request.
What are this agent's strengths and limitations?
- Two explicit paths (revive an existing mod vs. build from zero), each with matching skills and playbook guides
- Bundled MCP server for searching the 230+ entry co-op directory, engine detection from an install folder, and contribution validation
- Engine-family playbook mapping concrete tooling (UE4SS, Dumper-7, BepInEx, Harmony, xNVSE, REFramework, N64Recomp, PCSX2 PINE, Dolphin Memory Engine) to specific games
- Steam Spacewar relay means friends join without port forwarding and without extra purchases
- Explicit QA levels distinguish bot-tested releases from untested or player-reported entries
- Contributions flow back as human-reviewed draft PRs, so one agent's partial work seeds the next
- No finished mods are shipped here: you need the game, a coding agent and patience, and some engines take far longer than others
- Undocumented engines still fall back to low-level RE with Ghidra, REA or x64dbg, which is a steep skill barrier
- Online play depends on Steam: every participant must own and install the same game
- Documentation is English-heavy and leaves install specifics such as the Python version and dependency list unspecified
- Directory quality and project status depend on continued community submissions; many listed games are still in progress
How does this agent compare with similar options?
Complementary to Nucleus Co-op (Windows split screen) and PartyDeck (Linux/Steam Deck split screen), which it orchestrates over while adding online co-op and code-level work; complementary to a pure reverse-engineering MCP such as REA, where Co-op Anything decides what to do and REA looks inside the game.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Co-op Anything This agent | 37 · Major gaps | Agent plugin / skillFree + model costs | ★ 381 | 1d ago | C | Codex · Claude Code |
| REA: Reverse Engineer Anything with Agents | 72 · Some gaps | CLIFree | ★ 335k | today | TypeScript | Codex · Claude Code · Claude.ai |
| REA Reverse Engineering Workbench | 52 · Major gaps | CLIFree | ★ 65 | 1d ago | HTML | Codex · Claude Code |
| FastAgent | 46 · Major gaps | CLIFree + model costs | ★ 3.9k | 2d ago | Python | Codex · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
README states the agent asks before installing anything or using accounts and forbids touching anti-cheat or others' multiplayer, which supports user confirmation and external effects; however the repository only contains README, LICENSE and a directory-check workflow, with no code evidence of least-privilege design, data-flow disclosure, sensitive-data handling or rollback, so those score 1. Source attribution is explicit (credit mods built on, give fixes back), scoring 2.
README, workflow and directory structure are broadly self-consistent, but there is no substantive evidence of failure messages, error handling or dependency availability; the workflow only installs pyyaml and runs build_directory.py --check, a narrow dependency surface with no shown failure output, so failure_messages scores 0.
README covers audience (non-coders, users of Claude Code/Codex etc.) and scenarios (single-player co-op, split-screen, console emulators) well, scoring 2; capability boundaries and trigger precision are only described in prose without verifiable definitions, scoring 1 each; environment fit (Windows/Linux/Steam Deck, multiple engines) is described fairly fully, scoring 2.
Information architecture is clear (directory, FAQ, doc links), examples and FAQ are rich, and the MIT license is complete, each scoring 2; install notes give only an MCP add command with limited detail, scoring 1; naming stability, known limitations and maintenance responsibility appear only in scattered prose, scoring 1 each; no version number or changelog, scoring 0.
README promises one-click co-op packages and automated tests, but the repository itself contains neither those artifacts nor test evidence, so output usability, marginal value and cost-benefit rest on claims only, scoring 1 each.
README references multiple external game repositories and directory entries, giving some traceability, but the repository lacks corresponding evidence files, so cross-source corroboration and fact/inference separation are only partially supported, scoring 1 each.
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- The repository is mostly documentation and playbook; there is no reviewable code, tests or failure-handling evidence, so static assessment cannot confirm actual capability.
- README claims the agent asks before installing or using accounts, but no enforcement mechanism or permission boundary is implemented.
- Publisher identity is unverified by the FollowAgents registry; maintenance responsibility and update path are unclear.
- No version number or changelog, making behavior changes hard to track.