SkillAgentSearch skills...

worldbox-mcp

An MCP server that lets any AI agent (Claude Code, OpenCode, Codex, Cursor, ...) directly control the game WorldBox. Includes a BepInEx C# mod that exposes a local HTTP API to the game's internals, and a Python MCP server distributed on PyPI.

Install / Use

claude mcp add fullya99 -- npx -y github:fullya99/worldbox-mcp

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
šŸ”Œ

MCP Server

Model Context Protocol server

Quality Score

83/100

Supported Platforms

Claude Code
Claude Desktop
Cursor
OpenAI Codex

Our assessment of worldbox-mcp

worldbox-mcp scores 83/100 on our quality scale, 657th of 956 AI & Machine Learning skills we index.

Its MCP Server is 11 KB long, well organised into 17 sections with 11 code examples: a thorough specification that gives an agent plenty to work with.

It has 10 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
29/30
Structure
20/20
Description
15/15
Adoption
4/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated today, so worldbox-mcp is actively maintained.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 97/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

worldbox-mcp compared with similar skills

All 4 of these similar skills score higher than worldbox-mcp; compare them before choosing.

SkillScoreStarsUpdatedFormat
worldbox-mcp (this skill)by fullya998310todayMCP Server
claude-memby thedotmack100101.7ktodayCLAUDE.md
Agent-Reachby Panniantong10099.4k4d agoCLAUDE.md
Understand-Anythingby Egonex-AI10087.0k7d agoCLAUDE.md
headroomby headroomlabs-ai10076.0k1d agoCLAUDE.md

Frequently asked questions

How do I install worldbox-mcp?
Run claude mcp add fullya99 -- npx -y github:fullya99/worldbox-mcp. The install tabs above show the steps for each supported agent.
Which AI agents does worldbox-mcp work with?
It is written for Claude Code, Claude Desktop, Cursor and OpenAI Codex, as a MCP Server file. Other agents that read the same format can often use it too.
Is worldbox-mcp safe to use?
It is MIT-licensed and scores 97/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is worldbox-mcp still maintained?
The repository was last updated today, so worldbox-mcp is actively maintained.

worldbox-mcp

Give your AI agent god-mode in WorldBox. Spawn dragons. Burn continents. Run civilization speedruns. From a conversation.

CI License: MIT Code style: ruff MCP BepInEx

worldbox-mcp is a two-piece bridge that lets any MCP-compatible AI client (Claude Code, OpenCode, Codex, Cursor, Continue, ...) directly control the live game WorldBox.

<!-- gen-docs:begin total -->29<!-- gen-docs:end total --> MCP tools spanning discovery, action, observation, simulation control, multi-agent identity, and an inter-agent message bus. The agent observes the world, chooses what to do, acts on it, then observes the outcome, a full agentic loop running against an actual running game. Since v0.3, multiple AI agents can share the same world simultaneously with role-based permissions, fog-of-war, turn-taking, and direct messaging. See docs/multi-agent.md.


What it does

> You: "Build me a Roman empire surrounded by hostile dragons, fast-forward
        50 years of evolution, then tell me who's still alive."

> Agent:
    pause()
    list_actors() → finds 'human', 'dragon'
    list_tiles()  → finds 'soil_high', 'sand'
    paint_tile(soil_high, x=80, y=80, radius=10)       # founder territory
    spawn(human, x=80, y=80, count=15, adult=True)     # the empire
    spawn(dragon, x=200, y=80, count=4, adult=True)    # the threat
    set_speed("x5")
    resume()
    # ... waits ~10 seconds for 50 years of simulation ...
    screenshot()
    query_actors(race="human", alive=True)
    list_kingdoms()
    → "After 50 years, the Romans built 2 cities. The dragons
       razed the western one but the eastern capital Aletalda
       still stands with 47 citizens."

The whole exchange is real, examples/scenarios/ecology_smoke.py runs a stripped-down version of it as a smoke test.

How it works

ā”Œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”  MCP stdio  ā”Œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”  HTTP 127.0.0.1   ā”Œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”
│  AI client  ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā–ŗā”‚ worldbox-mcp     ā”œā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā–ŗā”‚ WorldBoxBridge      │
│ (any MCP)   │             │ (Python on PyPI) │                   │ BepInEx C# plugin   │
ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜             ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜                   │ inside worldbox.exe │
                                                                   ā””ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”€ā”˜
  • WorldBoxBridge: a BepInEx 5 plugin (C#) injected into WorldBox that exposes a local, auth-protected HTTP API to the game's internals.
  • worldbox-mcp: a Python MCP server distributed via PyPI (uvx worldbox-mcp) that translates MCP tool calls into HTTP requests against the mod.

The mod uses only reflection to access game types (no static binding) so it survives WorldBox updates as long as core class names stay stable. All asset ids (tile types, actor races, powers, world speeds) are discovered at runtime from the running build, so the catalog is always correct.

Quickstart

1. Install the in-game mod

Requires WorldBox installed (Steam) and Experimental Mode enabled in-game (Settings → Experimental Mode).

# Windows
iex (irm https://raw.githubusercontent.com/fullya99/worldbox-mcp/main/scripts/install-mod.ps1)
# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/fullya99/worldbox-mcp/main/scripts/install-mod.sh | bash

This downloads BepInEx + the latest WorldBoxBridge.dll, installs both into your WorldBox folder, and generates a per-install random auth token. Launch WorldBox once and look for this line in BepInEx/LogOutput.log:

[Info: WorldBoxBridge] listening on http://127.0.0.1:8723

2. Plug it into your AI client

The MCP server runs via uvx (zero install) thanks to the uv tooling.

<details> <summary><strong>Claude Code</strong>, one command</summary>
claude mcp add worldbox -- uvx worldbox-mcp
</details> <details> <summary><strong>OpenCode</strong></summary>
// ~/.config/opencode/config.json
{
  "mcp": {
    "worldbox": { "type": "local", "command": ["uvx", "worldbox-mcp"] }
  }
}
</details> <details> <summary><strong>Codex CLI</strong></summary>
# ~/.codex/config.toml
[mcp_servers.worldbox]
command = "uvx"
args = ["worldbox-mcp"]
</details> <details> <summary><strong>Cursor</strong></summary>
// .cursor/mcp.json
{
  "mcpServers": {
    "worldbox": { "command": "uvx", "args": ["worldbox-mcp"] }
  }
}
</details> <details> <summary><strong>Continue</strong></summary>
# ~/.continue/config.yaml
mcpServers:
  - name: worldbox
    command: uvx
    args: [worldbox-mcp]
</details>

Any other MCP-compatible client: see docs/install/manual.md. The server also supports a --http flag for Streamable HTTP transport.

3. Try it

In your agent client:

> Call worldbox_health to verify the bridge is connected.
> Now list_actors and pick a few interesting races, then build me
  a small ecology experiment near the center of the map.

Or paste the full god-mode ecology prompt from examples/prompts/godmode-ecology.md to exercise every tool category in one go.

The <!-- gen-docs:begin total -->29<!-- gen-docs:end total --> tools

| Category | Tools | What for | |---|---|---| | Meta | worldbox_health, worldbox_capabilities, worldbox_whoami, worldbox_session_info, worldbox_turn_advance, worldbox_objective_status | Liveness, introspection, multi-agent identity, turn rotation, scoreboard | | Discovery | worldbox_list_tiles, worldbox_list_actors, worldbox_list_powers, worldbox_list_speeds | Enumerate asset ids the running game exposes, ~20 tiles, ~320 actors, ~340 powers, 10 speeds on stock 0.51.x | | Action | worldbox_invoke_power, worldbox_spawn, worldbox_paint_tile | Modify the world: trigger powers, spawn creatures, paint terrain | | Read | worldbox_get_world_state, worldbox_get_tile, worldbox_list_kingdoms, worldbox_list_cities, worldbox_query_actors, worldbox_get_ui_state, worldbox_screenshot | Observe before deciding (fog-of-war filtered in multi-agent mode) | | Control | worldbox_pause, worldbox_resume, worldbox_dismiss_window, worldbox_set_speed, worldbox_generate_world, worldbox_save_world, worldbox_load_world | Simulation flow + world lifecycle | | Bus | worldbox_send_message, worldbox_recv_messages | Inter-agent messaging (1-to-1 or broadcast) for multi-agent sessions |

Each tool has rich docstrings the agent reads at startup to decide when to call what. Asset ids that don't exist trigger UNKNOWN_ASSET errors with Levenshtein did_you_mean suggestions drawn from the live catalog.

Full reference: docs/command-reference.md.

Design principles

  • Survives game updates. Every game type/method is resolved by reflection with caching and explicit warnings if a symbol disappears, a renamed class disables only the affected command, not the whole bridge. The mod also reports the WorldBox version and the SHA256 of Assembly-CSharp.dll in every /health so bug reports are traceable.
  • 100% coverage through 3 primitives. invoke_power alone covers ~270 actions because the in-game GodPower model unifies spawns, disasters, and toggles. spawn reaches the actor catalog directly for non-power creatures. paint_tile handles terrain. The discovery tools tell the agent what's valid right now in this build.
  • Local-only by design. The HTTP bridge binds only to 127.0.0.1, 0.0.0.0 is refused at startup. Auth is a per-install random token in BepInEx/config/WorldBoxBridge.cfg, or one bearer per agent in multi-agent mode via BepInEx/config/WorldBoxBridge.agents.json. Constant-time token comparison. No telemetry, no remote endpoints.
  • Multi-agent ready (v0.3). N AIs on one world via a session layer: per-agent role (god / faction_player / observer / narrator), permission gates, fog-of-war scoping, turn-based mode, bounded inter-agent message bus, scoreboard primitive. Four ready-to-customize presets (PvP / coop / hierarchical / sandbox) in examples/scenarios/multi-agent/.
  • Production-grade. Typed (mypy strict + C# nullable enabled), tested (xUnit for the mod, pytest for the server, both green without the game), formatted (ruff + csharpier), signed releases via release-please, automated CI/CD across Win/Linux/Mac Ɨ Py 3.11 to 3.13.

Compatibility

| WorldBox version | Unity | Mod version | Status | |---|---|---|---| | 0.51.2 | 2022.3.60f1 (Mono) | 0.3.x | āœ… multi-agent layer end-to-end (PvP smoke, fog-of-war, turn-based, bus) | | 0.51.2 | 2022.3.60f1 (Mono) | 0.2.x | āœ… generate/save/load world | | 0.51.2 | 2022.3.60f1 (Mono) | 0.1.x | āœ… baseline 20-tool surface |

A daily CI cron monitors WorldBox releases on Steam and opens a tracking issue when a new version ships, so update breakage gets caught the day-of. See docs/compatibility.md.

Architecture deep-dive

Contributing

Issues, PRs, new commands welcome. See CONTRIBUTING.md. Good first contributions: drive click_special_action (the last power delegate invoke_power cannot reach), or a get_actor(name_or_id) single-actor lookup.

License

MIT Ā© 2026 fullya99

worldbox-mcp is an unofficial community project and is not affiliated with or endorsed by the WorldBox developers.

Related Skills

View on GitHub
GitHub Stars10
CategoryAI
Updated5h ago
Forks1

Languages

C#

Trust signals

97/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 info
worldbox-mcp — MCP Server: Install & Safety Check | SkillAgent