Loading ToolBench

Reading the current corpus…

unity-resource-rag MCP server · D · ToolBench

unity-resource-rag

Asset-aware UI retrieval and blueprint execution tools for Unity projects. Exports project UI resources into a normalized catalog and applies real-asset UI blueprints through MCP-friendly custom tools.

2026-07-28+STDIO6 tools

54/100

Definition Quality61

Protocol Readiness35

SupportabilityNot measured-

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Server has 6 tools with mixed quality. Three tools (apply_ui_blueprint, index_project_resources, query_ui_asset_catalog) have reasonable schemas and descriptions visible in C# code. Three tools (extract_reference_layout, build_vector_index, search_catalog) are defined in Python with incomplete visibility, schemas inferred rather than explicitly registered. Naming is verb-forward but descriptions vary widely in completeness. Critical gaps: extract_reference_layout has 15+ parameters with minimal descriptions; build_vector_index lacks output schema documentation; error handling is present but recovery guidance is sparse. No tool annotations (readOnlyHint/destructiveHint) despite clear WRITE/READ_ONLY risk classifications.

Tools (6)

apply_ui_blueprintwritesource verified70/100

Validate or apply a structured UI blueprint. Supports canvas, container, prefab_instance, image, and tmp_text nodes plus layout groups, layout elements, and project-specific custom components.

build_vector_indexwrite50/100

Build a vector-searchable index from a Unity resource catalog.

extract_reference_layoutwriteauthsource verified38/100

Extract a UI layout from a reference image using vision analysis.

index_project_resourceswritesource verified75/100

Scan Unity UI-related project assets and export a normalized JSONL catalog plus optional previews for asset-aware UI retrieval.

query_ui_asset_catalogread onlysource verified70/100

Read the exported UI asset catalog as a paged tool result. Use this when the MCP client is better at tools than resources.

search_catalogread only50/100

Search the UI asset catalog using vector similarity.

Issues (5)

CRITICALtool-description ↗

extract_reference_layout has 15+ parameters with minimal or missing descriptions. Parameters like 'connection_preset', 'auth_mode', 'api_key_env', 'oauth_token_env', 'oauth_token_file', 'oauth_token_command', 'codex_auth_file', 'custom_base_url', 'gateway_auth_token_env', 'model', 'detail', 'max_image_dimension' lack actionable guidance. LLMs cannot infer valid values or when to use each auth variant.

extract_reference_layout

HIGHtool ↗

Three Python tools (extract_reference_layout, build_vector_index, search_catalog) lack visible explicit tool registration with MCP framework. Schemas are inferred from docstrings rather than formally declared. Cannot verify if input/output schemas are properly registered with the MCP server.

extract_reference_layoutbuild_vector_indexsearch_catalog

HIGHresponse-shaper ↗

No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., apply_ui_blueprint returns 'created' array but structure is undocumented; search_catalog returns results but field names/types unknown.

Recommendations

Add explicit descriptions to all 15+ parameters in extract_reference_layout. For each auth variant (api_key_env, oauth_token_env, oauth_token_file, oauth_token_command, codex_auth_file, custom_base_url, gateway_auth_token_env), explain when to use it, expected format, and example environment variable names.
Document output schemas for all tools. E.g., apply_ui_blueprint should specify: 'Returns {screenName: string, hasErrors: boolean, issues: [{severity: string, message: string}], rootInstanceId: number, rootName: string, created: [{nodeKind: string, nodeName: string, instanceId: number}]}'.
Add tool annotations to all tools. Mark apply_ui_blueprint and index_project_resources with destructiveHint=true; mark query_ui_asset_catalog and search_catalog with readOnlyHint=true; mark build_vector_index with destructiveHint=true (overwrites index).
Verify Python tools (extract_reference_layout, build_vector_index, search_catalog) are explicitly registered with MCP framework using proper tool registration API. Ensure schemas are formally declared, not inferred from docstrings.
Enhance error messages with recovery guidance. E.g., 'Blueprint validation failed: canvas node missing required property "width". Ensure all canvas nodes include width and height.' instead of generic 'Blueprint validation failed.'

Score history

Overall score trend

First recorded score · v2 rubric

54/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	54	2026-07-28+	v2