A web application for transforming audio samples using natural language processing and audio effects
FreqModa is a SvelteKit web application with a chat interface, not an MCP server. The source code shows NO MCP server implementation, no MCP protocol handlers, no tool registration via MCP, and no MCP transport. The three tools listed (processAudio, getAudioMetadata, healthCheck) are exposed via HTTP REST endpoints (/api/chat, /api/health) and invoked indirectly through OpenAI function calling, NOT through MCP. This is a critical misclassification. The evaluation below treats the hypothetical MCP exposure of these tools, but the server does not actually implement MCP. Tool descriptions are generic and lack LLM optimization. Parameter schemas are partially visible but lack type constraints and enums. No error handling guidance, no security annotation, and no output schemas documented.
Retrieve metadata for an audio file from the streaming engine
Check the health status of the streaming engine service
Process audio through the streaming engine with specified parameters including format, speed, reverse, volume, filters, and effects
NOT AN MCP SERVER: Source code is a SvelteKit web application with OpenAI function calling, not an MCP protocol implementation. No MCP tool registration, no MCP transport, no MCP request/response handling visible. Tools are HTTP REST endpoints, not MCP tools.
Parameter 'params' in processAudio is a free-form object with no schema, type constraints, or nested field descriptions. LLM cannot determine which fields are required, what types they accept, or what values are valid. Effect names (echo, chorus, flanger) should be enums but are described as strings.
Descriptions are generic and under-optimized for LLM reasoning. 'Retrieve metadata', 'Process audio', 'Check health' lack context on when to use each tool, what happens on success/failure, and what the output looks like. Descriptions fail to guide tool selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 32 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
No output schemas documented. LLM cannot determine what fields to expect in responses, forcing it to guess and parse unstructured output. No guidance on success/failure formats or error recovery.
No error handling guidance. Tool descriptions do not explain when calls fail, what errors to expect, or what the LLM should do next (retry, ask user, fallback). No error classification (retryable vs fatal).
Parameter 'audioName' in processAudio and getAudioMetadata accepts arbitrary strings with no validation guidance (case-sensitive? file extensions required? valid samples only?). LLM cannot determine what values are valid.
No security annotations. processAudio is marked 'WRITE' but no permission model, rate limits, or audit trail guidance documented. No mention of who can call this or what the consequences are.