MCP Interoperability

Store user settings, conversations, provider profiles, and credentials locally without leaking them into source control.

Safely create skills or tools and extend UnrealAgentTools itself using live registry capabilities, source tools/external coding agents, builds, and validation.

Add bounded, user-visible Unreal context that can be injected into model requests.

Implement a new raw LLM provider behind ILLMProvider without leaking vendor behavior into the host.

Add one small AICallable Unreal operation with bounded structured output.

Build compact inspect-first Blueprint graph capabilities and transactional edits suitable for LLM reasoning.

Diagnose connection, model discovery, streaming, malformed response, and tool-call problems at the provider boundary.

Integrate a process-oriented coding agent such as Codex CLI or Claude Code without pretending it is a raw LLM API.

Extend the built-in dockable Slate frontend while preserving host responsibilities and persistent conversations.

Classify tool risk, require appropriate approval, and make editor mutations transactional where practical.

Understand where code belongs and preserve the generic two-module architecture.

Create a new coherent Toolset Registry namespace/category and register AICallable operations.