MCP server for creating prompts in Ashra, a web data extraction platform
Single tool 'create_prompt' has a valid schema generated from Zod, but the description is minimal (8 words, well below the 10-char minimum for meaningful context). The schema itself is overly complex and deeply nested, making it difficult for LLMs to understand the expected structure. Parameter descriptions exist but are sparse. The tool performs a write operation (creates a prompt) but the description does not explicitly state this, violating the command-tool pattern. No error recovery guidance is provided. The schema nesting (object → properties → array → items → object → properties) is convoluted and would benefit from simplification or documentation of the expected data shape.
Create a single prompt in Ashra
Tool description is too brief (8 words: 'Create a single prompt in Ashra'). Does not explain WHEN to use it, what it returns, or that it modifies state. LLMs cannot determine selection criteria or side effects.
Schema parameter 'schema' is deeply nested (object → properties → array → items → object → properties) with no documentation of the expected structure. LLMs cannot infer what data shape is valid. Simplify or add a schema description explaining the expected JSON structure.
No error handling guidance. Tool calls external API (ashra.ai) but provides no recovery hints. If the API returns 401 (auth failure), 404 (URL not found), or 400 (invalid schema), the LLM has no guidance on what to do next.
API key (ASHRA_API_KEY) is injected server-side, which is correct. However, the tool description does not mention that authentication is required or what happens if the key is missing. LLMs may not understand why calls fail.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
Output schema (CreatePromptResponseSchema) is defined but not documented in the tool description. LLMs do not know what fields to expect in the response, making downstream planning difficult.