MCP gateway with argument-level policies, audit log and rule learning.
Single tool 'ping' with minimal definition. Tool has a description (good), but parameter description is in Russian ('Сообщение, которое сервер вернёт обратно'), making it inaccessible to English-speaking LLMs. No output schema documented. Tool is trivial (echo-only) and does not represent a production-grade MCP server. This is a spike/test fixture, not a real service. Naming is generic ('ping') and does not follow verb_noun pattern for clarity. No error handling, no pagination, no composition with other tools.
Echo back a message with the server's name
Parameter description in Russian ('Сообщение, которое сервер вернёт обратно'), LLMs trained on English cannot parse non-English descriptions. Descriptions must be in English for cross-language LLM compatibility.
No output schema documented. LLMs cannot predict what fields the response contains or plan downstream tool calls. Tool returns a string, but this is not formally declared.
Tool name 'ping' is generic and does not follow verb_noun pattern. 'echo_message' or 'echo_with_server_name' would be clearer and help LLMs infer intent from the name alone.
No error handling or recovery guidance. If the tool fails, LLMs have no actionable next step. Error responses should categorize failures as retryable, user-fixable, or fatal.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | <=2025-11-25 | v2 |
This is a spike/test fixture (spike/src/bin/echo-server.rs), not a production MCP server. The main Zastava gateway (crates/zastava-cli) is not evaluated here, only the test echo server is visible in the provided code.