Fleet-wide lethal-trifecta security scanner for Model Context Protocol (MCP) setups
This reference/test server has minimal production quality. All 4 tools have basic descriptions (10-60 chars, below the 194-char baseline) and schemas present but underdocumented. No parameter descriptions exist, LLMs cannot infer what 'path', 'url', or 'data' mean without context. No output schemas documented. Error handling is trivial (mock responses only). The server is designed as a test fixture, not a production tool, which explains the gaps. Naming is acceptable (verb_noun pattern), but parameter quality is poor.
Add two integers and return the sum.
Fetch and return the text content of any web URL.
Send an HTTP POST request to a configured webhook endpoint.
Read the contents of a file from the local filesystem.
No parameter descriptions. 'path', 'url', 'data', 'a', 'b' lack context. LLMs cannot infer whether 'path' is absolute/relative, whether 'url' must be HTTPS, or what format 'data' expects.
No output schemas documented. Callers cannot know what fields to expect from responses. The mock _text_result() returns {content: [{type, text}], isError}, but this is not declared in tool metadata.
Tool descriptions are too short (10-60 chars vs 194-char baseline). 'Read the contents of a file from the local filesystem' lacks WHEN to use it, prerequisites, or error guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
No input validation or constraint documentation. 'path' could be absolute, relative, or a symlink, no guidance. 'url' could be HTTP, HTTPS, or malformed, no format constraint.
post_webhook lacks security guidance. No mention of authentication, rate limits, or what happens if the webhook endpoint is unreachable. Agents cannot reason about failure modes.