How it works

The three moving parts — your editor, your AI client, and the Workbench cloud — and who does what.

Three parts make up the Workbench, and each has one job:

Unreal Editor 5.8  ──local MCP──▶  your AI client  ──remote MCP──▶  Unreal AI Workbench
(editor actions,                    (orchestration,                 (analysis, knowledge,
 asset inventory)                    prompts, approvals)             billing)

Your editor

Unreal Engine 5.8 ships a built-in MCP server inside the editor process. The connector plugin registers local tools on it — like the asset-inventory export — that operate on your project directly. This channel is local to your machine.

Your AI client

Claude Code, Cursor, opencode — the client is the orchestrator. It talks to both sides: local tools for editor actions, and the remote Workbench server for the heavy tools. When a tool needs your decision — approving a generated script, choosing to top up — the client is what asks you.

The Workbench cloud

The remote MCP server (mcp.spaghetti.gg) hosts the billed and research tools: log analysis, asset audits, script generation, documentation and pattern search. Compute for analysis runs here, on stored copies of your files — see Security and data. Every call is authenticated (OAuth), billed against your wallet (credits reserved before work, charged on success), and metered to your Usage page.

Why this split

Editor automation stays local and instant; expert reasoning — multi-megabyte log analysis, documentation-grounded research — runs server-side where the models and indexes live. Your client connects the two, which is why one prompt can chain them: analyze a log, search the docs for the top finding, then draft a reviewed fix script.

Next steps

Was this page helpful?