MCP server with RA-TLS attestation support for SGX TEE environments
Single tool 'fetch' has a basic schema and minimal description. The tool name is a verb (good), but the description is only 44 characters, below the 50-200 char baseline for LLM-optimized descriptions. The input schema is present with proper JSON Schema structure (type, required, properties), but lacks parameter-level descriptions beyond a bare 'URL to fetch'. No output schema is documented. No error handling guidance, no pagination support, and no indication of what fields the response contains. The tool is read-only (good security posture), but lacks the depth expected of production tools.
Fetches a website and returns its content
Tool description is only 44 characters ('Fetches a website and returns its content'). Baseline for LLM-optimized descriptions is 50 - 200 chars. Expand to explain WHEN to use this tool, what it returns, and any prerequisites (e.g., 'Fetches the full HTML content of a website via HTTP GET. Returns the response body as plain text. Useful for reading web pages, documentation, or public APIs. Follows redirects and includes a standard User-Agent header.').
Output schema is not documented. The code returns `list[types.TextContent | types.ImageContent | types.EmbeddedResource]`, but the tool definition does not declare this in the schema. LLMs cannot plan downstream operations without knowing what fields to expect. Add an explicit 'outputSchema' or document the response structure in the description.
Parameter 'url' lacks a description in the schema. The description field is present but empty. Add guidance: 'The full URL to fetch (e.g., https://example.com). Must be a valid HTTP or HTTPS URL. Relative URLs are not supported.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No error handling guidance. The code calls `response.raise_for_status()`, which will throw on 4xx/5xx, but the tool definition does not document what errors are possible (network timeout, invalid URL, 404, 403, etc.) or how the LLM should recover. Add error classification: 'Returns an error if the URL is invalid, unreachable, or returns a non-2xx status. Retry on network timeouts; ask the user to verify the URL for 404/403 errors.'
No result limits or pagination. If a website returns 10 MB of HTML, the entire response is returned to the LLM, potentially exhausting context. Add a note: 'Response is capped at 100 KB of text. Larger pages are truncated.' Consider adding a 'max_chars' parameter to let the LLM control response size.