Moderation API

@moderation check this user comment for policy violations: 'Hey everyone, check out my store at example.com for cheap deals! DM me for discount codes.'

The Moderation API evaluated the comment and flagged two policy violations: commercial spam (confidence: 0.92) and solicitation (confidence: 0.87). Recommended action: REJECT_CONTENT. The comment contains promotional language and a direct commercial link, which violates typical community guidelines around unsolicited advertising. No data was stored during this evaluation.

This example uses the ephemeral evaluation tool, which returns findings without persisting content. Useful for real-time checks before publishing. Note that confidence thresholds and policy definitions vary by your organization's configuration—what counts as 'spam' depends on the rules you've defined in your Moderation API project settings.

@moderation apply action 'suspend_author_7d' to user ID usr_abc123 for repeated spam violations

Action 'suspend_author_7d' successfully applied to user usr_abc123. The user account is now suspended for 7 days and cannot post new content or comments during this period. This action was logged in your moderation audit trail with timestamp 2025-01-15T18:42:33Z. The suspension will automatically lift on 2025-01-22T18:42:33Z unless manually revoked.

This demonstrates the execute-action tool, which applies real moderation consequences. Actions like suspensions and content rejections are often irreversible or require manual intervention to undo. Always verify the action ID and target before executing—use LIST_ACTIONS first to confirm available actions and their effects for your organization.

@moderation pull the last 20 items from review queue 'user-reports' and summarize which ones need immediate attention based on severity and report count

Retrieved 20 items from the 'user-reports' queue. Three require immediate attention: Item #4471 (hate speech, 12 user reports, confidence 0.95), Item #4468 (graphic violence, 8 reports, confidence 0.89), and Item #4455 (targeted harassment, 6 reports, confidence 0.91). The remaining 17 items are lower-priority spam or off-topic flags. All three high-priority items were flagged within the last 4 hours and involve different authors.

This synthesis example combines the list-queue-items tool with AI reasoning to triage a moderation backlog. Requires the paid review-queue add-on from Moderation API—if your organization hasn't purchased this feature, queue-related tools will return errors. The AI can help prioritize, but final moderation decisions should involve human judgment for context the API can't capture.

When you need content moderation before hiring moderators

A 6-person startup launching a user forum needs content filtering from day one but can't afford full-time moderators yet. This MCP is the right call if you're building moderation into your product workflow—evaluate-content-without-storage runs policy checks inline before posts go live, and execute-moderation-action can auto-reject or flag borderline cases. The wordlist tools let you tune filters as your community grows. The threshold: if you need human review queues for edge cases, you'll hit the paid add-on wall fast. If your volume is under 500 posts a day and your policies are clear-cut (profanity, spam, obvious hate speech), the free tier plus this MCP covers launch through your first few hundred users.

Filtering abusive support requests at scale

A 12-person SaaS company gets 200 support tickets a day, and 8% contain threats or abusive language that burn out the support team. This MCP wins if you want to auto-flag or auto-close the worst offenders before they hit your queue. Evaluate-content-without-storage runs on every inbound ticket; execute-moderation-action can suspend repeat offenders or route flagged tickets to a senior agent. The get-account-and-quota tool helps you track usage so you don't blow your API limit mid-month. The trade-off: if your ticket system already has moderation hooks (Zendesk, Intercom), you might not need the MCP layer—check if your platform's native integration is faster. If you're on a custom helpdesk or stitching together Slack and email, this MCP is the fastest path to sane filtering.

When review queues justify the paid add-on

A 20-person marketplace team processes 1,200 seller listings a week, mixing auto-approval with human review for edge cases. This MCP makes sense only if you're already paying for Moderation API's review-queue add-on—list-review-queue-items and get-review-queue let your ops team pull flagged listings into Switchy for batch decisions during daily triage. The execute-moderation-action tool closes the loop by applying bulk approvals or rejections back to the API. The boundary: if you're not paying for review queues, you lose half the MCP's value and should evaluate lighter tools like OpenAI's moderation endpoint first. If you are paying and your team lives in Switchy for ops work, this MCP turns moderation into a first-class workflow instead of a separate dashboard you context-switch to.