A Model Context Protocol server for monitoring shadow-cljs builds
shadow-cljs-mcp exposes a single tool with a moderately clear description but critically lacks parameter documentation and output schema specification. The tool definition is visible in index.js and properly registered via the MCP SDK, but fails to meet production-quality standards for parameter descriptions, schema documentation, and error handling guidance. The tool is READ_ONLY and has no destructive side effects, but the lack of structured output documentation and parameter constraints represent significant gaps from the 54 Agentic Tool Patterns baseline.
Get the status of the last shadow-cljs build including any warnings or errors. Call this after making edits to ClojureScript files to verify if the build succeeded or failed.
Input schema is empty, tool declares no parameters but provides no documentation of what the tool actually returns or how to interpret the response
Output schema is completely undocumented, LLMs have no way to know which fields to expect from the response (e.g., is 'status' a string enum or a boolean? What does 'websocket_connected' mean? What is the structure of 'report'?)
Tool description lacks critical context: When should the LLM call this vs. other monitoring approaches? What are the possible status values? What do 'warnings' and 'report' contain? Current description reads more as a usage note than a complete docstring.
No error handling guidance, if WebSocket is disconnected, the tool returns a default 'unknown' status with no recovery instructions. LLMs receive no hint that they should retry, check server status, or ask the user to verify the shadow-cljs server is running.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 32 | - | v1 |
Response fields like 'warnings', 'log', 'report', 'active', 'resources', 'compiled', 'duration' are returned without type specification or description. An LLM cannot reason about what these fields mean or how to use them in downstream operations.
Tool description does not clarify that this is a READ_ONLY, non-destructive operation. While the risk is correctly marked as READ_ONLY, the description should reinforce that this is safe to call repeatedly without side effects.