A MCP server to access any documentation indexed with RagRabbit
RagRabbit exposes 2 tools with minimal schema documentation and very sparse descriptions. The 'rag' tool has a basic description ('Search the knowledge base') that falls below the 50-character minimum for LLM clarity. The 'deploy' tool has similarly terse documentation. Neither tool has visible input schema definitions in the provided source code, only parameter names and types are inferred from a package.json file and a React component reference. No output schemas are documented. The server is a monorepo built on Next.js/React with MCP-SDK, but tool definitions appear embedded in application code rather than properly registered MCP tool handlers. Critical issues: (1) Tool definitions are inferred from filenames and comments, not explicitly registered; (2) Input schemas lack formal JSON Schema structure with proper type and description fields; (3) Descriptions are too short to guide LLM tool selection; (4) No error handling, recovery guidance, or output documentation.
Deploy repository to vercel
Search the knowledge base
Tool definitions inferred from code comments and file paths, not explicitly registered via MCP tool handler. No evidence of proper MCP ToolDefinition objects with name, description, and inputSchema fields.
Input schemas incomplete or missing. Parameters are listed (query, language for 'rag'; repositoryName for 'deploy') but no formal JSON Schema with type definitions, required field declarations, or constraints is visible in source.
Tool descriptions are too short (9 - 26 characters) to guide LLM selection. 'Search the knowledge base' lacks context for WHEN to use it vs alternatives, WHAT it returns, or HOW to interpret results.
'deploy' tool description ('Deploy repository to vercel') omits critical details: does it require authentication? What happens on failure? Is this idempotent? Can it be safely retried?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 36 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No output schemas documented. LLM cannot plan downstream operations or extract return values. 'rag' likely returns search results, but structure is unknown. 'deploy' return shape is undocumented.
Parameter 'language' in 'rag' tool is undocumented. Does it accept BCP 47 codes (en-US), ISO 639-1 (en), or natural language (English)? No enum, pattern, or format constraint.
No error handling or recovery guidance. If 'deploy' fails (invalid repository, auth error, network timeout), LLM receives no guidance on whether to retry, ask user, or escalate.
'deploy' is a destructive/state-modifying operation (Risk: WRITE) but has no confirmation step, dry-run option, or warning in description. Agent may trigger production deployments without user approval.