Frequently asked questions
Do I need an API key?
No. ClaudusBridge never asks for an API key and never bills you per token. You sign in with your own AI account through that provider's normal OAuth login flow — the same login you'd use in the official app. The plugin just drives the agent; your provider handles auth and model calls.
Which providers and models can I use?
Three providers: Claude (Anthropic), ChatGPT/Codex (OpenAI), and Gemini (Google). The embedded agent is a Claude Code fork (CCAG) that also drives ChatGPT/Codex and Gemini in the same session, so you pick whichever account you have. The available models are whatever your signed-in account has access to from that provider.
Can I use external MCP clients instead of the in-editor terminal?
Yes. The MCP server is a standard JSON-RPC 2.0 Streamable-HTTP endpoint at http://127.0.0.1:3000/mcp, so any MCP-capable client can connect and drive the editor. ClaudusBridge can write the connection config for common clients, including Claude Desktop / Claude Code, Cursor, Windsurf, VS Code, Gemini, and Codex. Point the client at the endpoint (or use the generated config) and the editor's 900+ tools become available there.
Is it local? Does it work offline?
It is 100% local in the sense that the plugin, the agent bundle, and the MCP server all run on your machine — nothing is routed through Galidar's servers. The only traffic that leaves your machine is your chosen provider's own login and model API calls. So it is local, but not fully offline: the AI provider's calls still need internet access, just like the official Claude/ChatGPT/Gemini apps.
Why do I need Node.js?
The embedded agent ships as a self-contained Node bundle inside the plugin (Resources/Agent/cli.js + cli-node.js). ClaudusBridge runs it with your system Node.js install, so a working node on your PATH is required. There's no separate agent installer — the bundle is already in the plugin.
What can the agent actually do in the editor?
A lot. Through 900+ MCP tools across ~65 toolsets, the agent can work with Actors and the World, Blueprints, Materials, Niagara, UMG/Widgets, Sequencer, Assets, Levels, and Textures — plus native plugin toolsets for Multi-User, Assets, Levels, Images, and Chess. You can also author your own tools by subclassing UClaudusBridgeToolLibrary; they auto-register and show up alongside the built-ins.
How is this different from a built-in chatbot?
A chatbot only talks. ClaudusBridge's agent acts. It drives the editor through real MCP tools generated from Unreal's reflection system — creating and editing assets, wiring Blueprints, building materials, setting up scenes — and runs in a genuine in-editor terminal (real PTY) or over the Agent Client Protocol (ACP) for programmatic control. It can read editor state, call tools, and apply changes, not just describe them.
What's Multi-User mode?
ClaudusBridge integrates with Unreal's Multi-User (Concert) system. The agent and a human can join the same Concert session as separate participants, each with their own desktop presence avatar, reusing Unreal's existing MultiUser client (auto-configured at startup — no engine fork, no second client). This lets you collaborate live with the agent in a shared session.
Which Unreal Engine version is supported?
Unreal Engine 5.8, on Windows, macOS, and Linux. A separate legacy build for UE 5.7 (the "Neural Cortex" generation) exists, but its feature set differs from this 5.8 generation and is documented separately — this documentation describes 5.8.
How do I run the server on a different port?
Set ServerPortNumber in UClaudusBridgeSettings, or pass -ClaudusBridgePort=N on launch. You can confirm the server is up by hitting GET /health, which reports the status, port, and MCP path. Other server behavior — the MCP path (ServerUrlPath), tool-search vs. eager mode (bEnableToolSearch), and auto-start (bAutoStartServer) — is also configured in the settings.
Where can I get help?
Join the Discord, visit galidar.com, or find the plugin on Fab.