UEMCP has 16 tools with mixed quality. Most tools have descriptions and schemas present, but critical issues emerge: (1) Naming inconsistency, tools use both verb_noun (viewport_mode, actor_delete) and noun_verb (asset_list, level_actors), creating LLM disambiguation burden. (2) Parameter descriptions are inconsistent, some parameters lack descriptions entirely (e.g., batch_spawn's 'actors' array has no item-level schema details; viewport_fit's 'padding' has no range guidance). (3) Output schemas are completely undocumented, no tool documents what fields it returns, forcing LLMs to guess at response structure for chaining. (4) Error handling is absent, no tool documents failure modes or recovery paths. (5) The server infers tool definitions from Python implementations via ManifestGenerator, and while schemas ARE visible in the rubric data provided, the actual code shows only partial type extraction logic. This falls into the 'inferred registration' category where confidence is reduced. Overall, this is a competent but incomplete implementation landing squarely in the C-/D+ range.

• Document output schemas for all 16 tools. For each tool, specify: (1) response structure (object fields, types), (2) pagination if applicable (total_count, next_cursor), (3) chaining IDs (e.g., batch_spawn must return spawned actor IDs for downstream actor_delete calls). See pattern:tool-chain and pattern:paginated-result.
• Standardize naming to consistent verb_noun convention (e.g., list_assets, list_level_actors, list_level_outliner instead of asset_list, level_actors, level_outliner). This reduces LLM disambiguation overhead.
• Add recovery-guide error handling to all tools. For each tool, document: (1) what can fail (e.g., 'actor not found', 'invalid socket name'), (2) retryability (is error transient?), (3) recovery step (e.g., 'call level_actors to find valid names', 'verify socket name via asset_info'). See pattern:recovery-guide.
• Add explicit enum constraints for all mode/choice parameters. Replace 'Operation mode' descriptions with actual enum list: viewport_mode modes (e.g., 'wireframe', 'lit', 'unlit'), viewport_render_mode modes, etc. Use JSON Schema 'enum' field and reiterate in description.
• For batch_spawn, expand 'actors' array schema to document item structure. Example: 'items object with properties: {blueprintPath (string, required), location (array [X,Y,Z], optional), rotation (array [Roll,Pitch,Yaw], optional), name (string, optional)}'. See pattern:tool-description.
• Add confirmation-request pattern to destructive tools (actor_delete, asset_import, batch_spawn with many actors). Introduce a 'confirm: true' parameter or a separate 'preview_batch_spawn' tool that returns what WOULD be spawned without modifying state.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry