For Agents
Score user-generated text for toxicity, insult, threat, and related attributes, and submit human-labeled feedback to improve the model. Lets agents triage harmful content in moderation pipelines.
Get started with Perspective Comment Analyzer API in minutes using your preferred integration method.
# Add to your MCP client config (Claude Desktop, Cursor, Windsurf)
{
"jentic": {
"url": "https://api.jentic.com/mcp",
"auth": "oauth"
}
}
# Then ask your agent:
"score comment for toxicity"
# → Jentic returns the GET /events tool with parameter schema, agent executes.What an agent can do with Perspective Comment Analyzer API API.
Score a comment on TOXICITY, SEVERE_TOXICITY, INSULT, THREAT, and PROFANITY attributes
Request span-level annotations for parts of a comment that triggered each score
Submit moderator-corrected scores back to the model via comments:suggestscore
Score text in multiple supported languages with a single request
GET STARTED
Use for: I need to score a comment for toxicity, Check whether a message looks like an insult or threat, Analyze a forum post for severe toxicity, Submit a moderator's score back to Perspective for training
Not supported: Does not block, delete, or rewrite content, identify users, or detect spam beyond its trained attributes — use for scoring text on toxicity-related attributes only.
The Perspective Comment Analyzer API scores user-generated text on attributes such as toxicity, severe toxicity, insult, profanity, and threat. Built by Jigsaw and Google's Counter Abuse Technology team, it powers comment moderation for newsrooms, forums, and chat apps. Callers submit a comment and receive per-attribute probability scores and span-level annotations, plus a suggestscore endpoint to send labels back into the model's training pipeline.
Filter or route content based on per-attribute probability thresholds
Patterns agents use Perspective Comment Analyzer API API for, with concrete tasks.
★ Pre-Publish Comment Moderation
Score every new comment with the analyze endpoint before it is shown to the public. Comments with toxicity probability above a threshold are routed to a human moderator queue, while clearly safe comments are published immediately. This lets community sites keep submission latency low without exposing readers to abuse.
Call comments:analyze with the comment text and the TOXICITY attribute, and route any comment with summaryScore.value > 0.7 to manual review
Real-Time Chat Filtering
Score chat messages as they are sent so the platform can soft-block or flag harmful content before it reaches recipients. Perspective returns per-attribute probabilities for TOXICITY, INSULT, THREAT, and PROFANITY in a single call, letting product teams build a tiered response (warn, hide, escalate) without retraining their own models.
Score every inbound chat message for INSULT and THREAT and hide messages where either summaryScore exceeds 0.85
Active Learning Feedback Loop
When human moderators override an automated decision, send their corrected score back to Perspective via comments:suggestscore. This signal is used by Jigsaw to improve future model versions on the attributes that matter most to your community, which keeps moderation quality high as language norms shift over time.
When a moderator marks an automated TOXICITY=0.9 decision as incorrect, call comments:suggestscore with the moderator's score of 0.1
Agent-Mediated Trust And Safety
An AI agent built into a community platform can call Perspective via Jentic to triage every report or new post, escalating only borderline cases to humans. Because Jentic returns the analyze schema, the agent can ask for the exact attributes it cares about and act on the structured probabilities directly.
Use Jentic to call comments:analyze for each reported post requesting attributes TOXICITY, SEVERE_TOXICITY, INSULT, THREAT, and surface any post where any attribute exceeds 0.8
2 endpoints — the perspective comment analyzer api scores user-generated text on attributes such as toxicity, severe toxicity, insult, profanity, and threat.
METHOD
PATH
DESCRIPTION
/v1alpha1/comments:analyze
Score a comment on the requested attributes
/v1alpha1/comments:suggestscore
Submit a corrected score for a previously analyzed comment
/v1alpha1/comments:analyze
Score a comment on the requested attributes
/v1alpha1/comments:suggestscore
Submit a corrected score for a previously analyzed comment
Three things that make agents converge on Jentic-routed access.
Credential isolation
Perspective API keys and Google OAuth credentials are stored encrypted in the Jentic vault. Agents receive only the scoped credential needed to call comments:analyze, and rotation is handled via vault update rather than code change.
Intent-based discovery
Agents search Jentic with intents like 'score comment for toxicity' and Jentic returns the comments:analyze operation with its request schema and supported attributes.
Time to first call
Direct integration with Perspective: 1-2 days for credential setup, attribute selection, and rate-limit handling. Through Jentic: under 30 minutes — search, load schema, execute.
Alternatives and complements available in the Jentic catalogue.
Cloud Search API
Indexes the moderated content corpus that Perspective scores upstream
Use Cloud Search to expose moderated content for retrieval; use Perspective to filter what gets indexed
Compute Engine API
Runs the moderation worker fleet that calls Perspective
Provision Compute Engine workers when you need to score very high volumes of content with parallel calls
Cloud Tasks API
Throttles outbound calls to Perspective so you stay under its QPS quota
Use Cloud Tasks to rate-limit your scoring calls to Perspective during traffic spikes
Specific to using Perspective Comment Analyzer API API through Jentic.
What authentication does the Perspective Comment Analyzer API use?
It uses Google OAuth 2.0 with the cloud-platform scope, and Perspective additionally accepts API keys at the request level. Through Jentic the credential is stored encrypted in the Jentic vault and the agent only sees scoped tokens at execution time.
Which attributes can the API score a comment on?
Common attributes include TOXICITY, SEVERE_TOXICITY, INSULT, IDENTITY_ATTACK, THREAT, and PROFANITY, with additional experimental attributes available for English text. The full list is supplied per-request via requestedAttributes in the comments:analyze body.
What are the rate limits for the Perspective API?
Perspective enforces a default per-project quota of one query per second, with higher limits available on request through the Perspective developer site. Batch your moderation calls and cache scores for repeated text to stay under the QPS ceiling.
How do I score a comment through Jentic?
Search Jentic for 'score comment for toxicity', load the comments:analyze schema, and execute the call against /v1alpha1/comments:analyze with the comment text and a requestedAttributes map. Jentic returns the schema so the agent fills only the attributes it cares about.
Is the Perspective API free?
Yes. Perspective is provided free of charge by Jigsaw within the published quotas. Higher-volume access requires a request through the Perspective developer portal.
Can I submit moderator decisions back to improve the model?
Yes. Call /v1alpha1/comments:suggestscore with the original comment, the requested attributes, and the corrected score from your moderator. Jigsaw uses these signals to retrain and improve future model versions.