Tiny, zero-dependency, pluggable telemetry SDK for MCP widgets (Claude MCP Apps, ChatGPT Apps SDK, mcp-ui). Capture lifecycle, custom, and error events and forward them to any analytics destination.
mcp-signal defines a single tool (record_signal) with a complete input schema and annotations. However, the tool is explicitly marked 'Internal: receives batched widget telemetry from the mcp-signal SDK. App-only; not intended for model use.' This is a telemetry sink, not an agent-facing tool. The schema has critical gaps: the 'events' array items are typed as bare objects with no property definitions, 'sdk' is an untyped object, and 'sentAt' lacks format specification. Parameter descriptions are minimal. The tool is not designed for LLM interaction, it's a bridge transport endpoint for widget-to-server communication. Scoring reflects that this is infrastructure, not a user-facing tool suite.
Internal: receives batched widget telemetry from the mcp-signal SDK. App-only; not intended for model use.
Input schema for 'events' array items is untyped (type: 'object' with no properties). LLMs cannot infer what fields each event should contain.
Parameter 'sdk' is an untyped object with no description. Unclear what structure or fields are expected.
Parameter 'sentAt' lacks format specification (ISO 8601? Unix timestamp?). No description provided.
Tool is explicitly marked 'not intended for model use', this is a telemetry sink for widget-to-server communication, not an agent-facing tool. The MCP server is infrastructure, not a tool suite.
No output schema documented. Unclear what the tool returns or whether it succeeds/fails.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |