MCP server for interacting with Crux applications via WebAssembly components
This MCP server has significant definition quality gaps across all tools. While tool names follow verb_noun conventions (schema, update, resolve, view), descriptions are minimal (12-34 characters), falling well below the 50-200 character baseline for A-grade tools. Input schemas are present but severely underspecified: three tools (schema, view, resolve effect_id) have no parameter descriptions whatsoever. The 'update' tool accepts a bare 'data' string with only one-line description, providing no guidance on format, constraints, or expected structure. Output schemas are completely undocumented, LLMs have no visibility into what these tools return, making downstream tool composition impossible. Risk designations (READ_ONLY, WRITE) are noted but not exposed in tool definitions. No error handling guidance exists. No pagination support. No field naming alignment between tools (no chaining IDs returned). The server appears to be a proof-of-concept rather than a production-ready MCP implementation.
Calls the resolve function in a Crux app
Gets the schema of a Crux app
Calls the update function in a Crux app
Gets the ViewModel by calling the view function in a Crux app
Missing parameter descriptions on critical inputs. The 'schema' tool has {} input schema with no description text. The 'view' tool has {} input with no description. The 'resolve' tool has 'effect_id' parameter with description '12' (under 20 chars) and 'data' with minimal guidance.
No output schemas documented. Tools return Result<Vec<u8>> (opaque byte strings), but LLMs have zero visibility into structure, field names, or content type. The 'schema' tool presumably returns a JSON schema, but the response format is undocumented, forcing LLMs to guess or parse raw bytes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 33 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | - | v1 |
Tool descriptions are too brief (5-34 chars vs 50-200 baseline). 'Gets the schema of a Crux app' (31 chars) lacks context on WHY or WHEN to call it, what it returns, or whether it's a prerequisite for other tools. 'Calls the update function in a Crux app' (40 chars) tells the LLM what to call but not what 'update' does semantically or what data structure the function expects.
No input validation guidance or constraints. The 'update' and 'resolve' tools accept 'data' as a string with zero specification of format (JSON? Binary? Plaintext?), encoding, max length, or structure. An LLM will guess and frequently submit invalid payloads.
No error handling or recovery guidance. The MCP handler (handler.rs) returns CallToolError with a bare error message on failure, but the tool definitions include no guidance on what errors are possible, whether they are retryable, or what the LLM should do next.
No tool composition guidance. There is no documentation of dependencies (e.g., 'call schema() first to understand the app structure before calling update()') or chaining IDs. If 'schema' returns a set of available effects, and 'resolve' accepts an effect_id, this linkage must be explicit in the definitions so LLMs can plan multi-step sequences.
Missing idempotency guidance. The 'update' and 'resolve' tools are marked WRITE but no mention of whether they are idempotent. If an agent retries a failed update call, will it duplicate the effect? Agents retry on ambiguous failures, non-idempotent tools risk duplicate side effects.'