Authenticate with the Observe.AI API using a bearer token obtained from your account credentials against the Observe.AI API base URL
POST call metadata (audio URL or transcript text, agent ID, call timestamp) to the Observe.AI ingestion endpoint to submit a call for analysis
Poll the interaction status endpoint with the returned interaction_id until status transitions from processing to completed
GET the completed interaction record to retrieve the automated scorecard — each scorecard item contains question text, AI-assigned answer, confidence score, and flagged moments
GET /agents/{agentId}/scorecard-summary with a date range to aggregate QA scores per agent for performance reports
Use the Moments API to retrieve timestamped highlights (e.g., compliance phrases spoken or missed) and export them alongside the transcript for QA review workflows
Known gotchas
Observe.AI's automated scoring uses a combination of ML, keyword spotting, and rules — scorecard items that rely on semantic understanding may have lower confidence on short or noisy calls; filter results below a confidence threshold for manual review
PII redaction is applied to transcripts before returning via API; audio content containing sensitive data may be redacted with placeholder tokens — downstream systems consuming transcript text must handle redaction markers
100% of interactions are auto-scored but the API enforces per-account rate limits on ingestion; large batch uploads of historical calls should be rate-controlled to avoid queuing delays that push status polling times into hours
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp