Model Context Protocol server for prompt template operations with AI integration, workspace analysis, and LLM provider proxying
Vibes MCP CLI has 8 tools with complete JSON Schema definitions and descriptions, which is a strong foundation. However, there are systematic issues that prevent a higher score: (1) Parameter descriptions are minimal or missing contextual guidance on formats/constraints; (2) No output schemas documented for any tool, LLMs cannot plan downstream calls; (3) Error handling and recovery guidance absent from descriptions; (4) Tool names lack consistency in verb-noun patterning (e.g., 'suggest_templates' is good, but 'detect_context' and 'workspace_analysis' are vague about return values); (5) Parameter defaults are present but some lack clear justification; (6) No evidence of pagination support on tools that return lists (get_history, suggest_templates, template_stats). The schemas themselves are well-formed with types and enums where appropriate, and all tools have descriptions, but descriptions are surface-level (avg ~120 chars, below the production baseline of 194 chars) and lack LLM-optimization guidance on when to use each tool.
Detect current workspace context (language, framework, dependencies)
Generate a prompt from a template with parameters and context
Retrieve prompt generation history with filtering
Perform comprehensive quality check on templates or generated content
Suggest relevant templates based on context
Get statistics about template usage and performance
Validate a prompt template or generated content
No output schemas documented for any tool. LLMs cannot infer return value structure and must guess what fields are available for downstream tool chaining. This violates the pattern:tool and pattern:response-shaper requirements.
Parameter descriptions lack guidance on expected formats, ranges, and constraints. E.g., 'start_date' and 'end_date' in get_history say 'ISO 8601' but do not specify examples or whether times are required. 'check_types' in quality_check defines enum values but description does not explain what each check does. This forces LLMs to guess.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 58 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 28 | - | v1 |
Analyze workspace for prompt optimization opportunities
Tool 'detect_context' and 'workspace_analysis' have vague names and descriptions about what they return. 'detect_context' says 'Detect current workspace context (language, framework, dependencies)', it is unclear whether it returns a single object or an array. No indication of return format inhibits agent planning.
No error handling guidance in any tool description. Tools like 'validate_template' and 'quality_check' do not explain what happens on invalid input, what error codes might be returned, or how LLMs should recover. Per pattern:recovery-guide, error descriptions must be actionable.
get_history and suggest_templates accept unbounded 'max_results' and have no documented result limits. Per mxe:enforce-result-limits, large result sets blow the context window and degrade LLM reasoning. Should cap at 20 - 50 items and document the limit in the description.
No pagination support visible on list-returning tools. get_history accepts 'limit' but no cursor, offset, or next_cursor return. suggest_templates returns up to max_results but no total count. Per pattern:paginated-result, agents need pagination metadata to fetch large result sets without context overflow.