Universal AI notification MCP server — send notifications to Slack, Google Chat, and Discord via webhooks
Three tools with complete schemas and descriptions. Naming follows verb_noun pattern (notify, notify_all, ping_webhook). Descriptions are detailed (150-200 chars) and explain WHEN to use each tool. However, parameter descriptions in notify_all.targets lack detail, the nested object properties have no descriptions, forcing LLMs to infer meaning. Error handling is present but generic (success/error boolean). Output schema is not documented, callers must infer the response structure from code. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite notify and notify_all being destructive write operations.
Send a notification to Google Chat, Slack, or Discord via webhook. Use this when an AI agent completes a task, encounters an error, or needs to alert a human. Returns success/failure so the agent can decide what to do next. Supports markdown formatting (wrap text in **bold**, *italic*, `code`). Set retries to auto-retry on transient failures.
Send a notification to multiple platforms simultaneously. Use this to alert different teams/channels at once. Each target can have its own platform, webhook, and formatting options.
Test a webhook URL to verify it's reachable and valid. Use this before sending important notifications to ensure the connection works. Returns the platform type and connection status.
notify_all.targets array items lack parameter descriptions. Nested properties (platform, webhook_url, message, title, status) have no descriptions, forcing LLMs to infer meaning from names alone.
Output schema not documented. Tools return {success, platform, message/error} but callers must infer structure from code. LLMs cannot plan downstream actions without knowing response fields.
Destructive operations (notify, notify_all) lack tool annotations. No destructiveHint or idempotentHint to signal to clients that these calls have side effects and may not be safely retried.
Error messages are generic. 'Invalid webhook_url' does not guide recovery. Should suggest: 'Invalid webhook_url. Must be HTTPS. Get one from Slack Incoming Webhooks, Google Chat Space Webhooks, or Discord Channel Webhooks.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
notify_all does not document per-target success/failure. If one target fails, does the entire call fail or return partial results? Ambiguous behavior forces LLMs to guess.