A modular AI agent CLI with MCP server support, desktop UI, and tool execution capabilities
Somnia's MCP server exposes only 2 tools with minimal definitions. Both tools have descriptions (positive), but parameter documentation is severely lacking. The 'echo' tool accepts additionalProperties=true with no parameter descriptions, violating the requirement that every parameter needs a description. The 'server_info' tool has no parameters but its output schema is undocumented. Neither tool has actionable descriptions that guide LLM selection or explain when to use them. Tool names are generic ('echo', 'server_info') and lack action verbs. No error handling guidance, no output schema documentation, and no composition patterns evident. This is a minimal reference implementation, not production-grade.
Echo the provided arguments back as formatted JSON text.
Return basic runtime information for stdio MCP connectivity checks.
Parameter 'message' in 'echo' tool lacks description. Schema declares additionalProperties=true but provides no guidance on what additional properties are valid or what they do.
No output schema documented for either tool. LLMs cannot plan downstream calls or extract structured data when return types are undocumented.
Tool descriptions are too generic and lack context for LLM selection. 'Echo the provided arguments back as formatted JSON text' does not explain WHEN to use this tool or what problem it solves.
Tool names lack action verbs. 'echo' and 'server_info' are generic; should be 'echo_message' and 'get_server_info' to signal intent clearly.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
No error handling guidance. If 'echo' fails or 'server_info' times out, the LLM has no recovery path documented.