System-tray MCP gateway that toggles MCP servers for all CLI agents at once
Patchbay exposes 4 gateway meta-tools for MCP server lifecycle management. All tools lack input schemas entirely (schema score = 0 for all). Descriptions are present but minimal (10-20 chars), failing the 20-char floor for description scoring. No parameter documentation visible. Tools are namespaced correctly (patchbay__*) but lack the structured definitions required for LLM-safe invocation. This is a gateway/infrastructure tool, not a user-facing agent tool, which explains the minimal schema investment, but MCP spec requires formal tool registration with schemas regardless of purpose.
Gateway-owned meta tool to add a new jack (MCP server)
Gateway-owned meta tool to list all configured jacks (MCP servers)
Gateway-owned meta tool to remove a jack (MCP server)
Gateway-owned meta tool to toggle a jack (MCP server) on or off
All 4 tools lack input schemas entirely. MCP spec requires JSON Schema for every tool's inputSchema field. Without schemas, LLMs cannot validate parameters, and clients cannot discover what arguments each tool accepts.
Tool descriptions are 10 - 20 characters (e.g., 'Gateway-owned meta tool to add a new jack (MCP server)'). Rubric baseline is 34 - 392 chars; descriptions under 20 chars score 0. These are too terse to guide LLM selection or explain when to call each tool vs. alternatives.
No parameter descriptions visible in source. Even if add_jack accepts a 'jack_name' or 'config' parameter, the LLM has no guidance on format, constraints, or dependencies. This violates the pattern:tool-description requirement that every parameter must have a description.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | <=2025-11-25 | v2 |
No output schema documented. Clients and LLMs cannot predict what fields list_jacks returns (e.g., does it return {jacks: [{name, status, pid}]} or a flat array?). This breaks tool chaining and forces agents to guess.
No error handling guidance. If add_jack fails (e.g., jack name already exists, invalid config), the tool provides no recovery hint. Agents cannot distinguish retryable errors from user-fixable ones.