MCP server for URKit mod development, providing tools to inspect project configuration, build mods, deploy artifacts, read logs, and interact with the URKit runtime via DevBridge
URKit MCP has well-structured tool definitions with complete input/output schemas and proper JSON Schema validation. All 7 tools have descriptions and most parameters are typed with constraints (minLength, maxLength, minimum, maximum). Tool annotations (readOnlyHint) are correctly applied. However, parameter descriptions are entirely absent, the schema declares constraints but provides no guidance text explaining what each parameter controls or when to use it. Output schemas are documented but lack descriptions for response fields. Error handling guidance is not visible in the tool definitions. The server demonstrates solid schema discipline but falls short on the LLM-facing documentation that drives correct tool selection and parameter usage.
Configure, build, and verify deployment of the URKit mod with a declared CMake preset.
Copy an existing preset build artifact to the manifest-owned Mods directory.
List runtime tests exported by loaded URKit mods.
Inspect the configured generated URKit mod project and deployment target.
Read a bounded tail of URKit_logs.log from the configured game directory.
Run one exported URKit runtime test on the Unity main thread.
Read backend, capability, process, and current scene state from URKit DevBridge.
No parameter descriptions. All 7 tools declare input parameters with type constraints (minLength, maxLength, minimum, maximum) but provide zero descriptive text. LLMs cannot infer what 'preset' means, when to use 'maximum_lines', or what values are valid beyond the schema constraints. This violates the critical pattern that every parameter needs a description explaining what it controls.
Output schema fields lack descriptions. Tools return structured objects (e.g., project_info returns project_root, project_name, sdk_version, backend, etc.) but provide no text explaining what each field represents or how to use it downstream. LLMs cannot reason about which fields to extract for follow-up calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2025-06-18+ | v2 |
runtime_status has no output schema. The tool is declared with inputSchema but outputSchema is missing entirely. LLMs cannot predict what fields will be returned, breaking downstream planning and field extraction.
list_runtime_tests has no output schema. Similar to runtime_status, the response structure is undocumented. LLMs cannot know whether the result is an array, object, or paginated response.
No error handling guidance in tool definitions. Tool descriptions do not explain what errors can occur, how to recover, or what the LLM should do if a call fails. For example, build_mod could fail if the preset does not exist, but there is no guidance on recovery.