Morphic CI intercepts MCP JSON-RPC payloads, generates prompt variations, and calculates F1 consistency scores. If your agent drifts, the deployment gets blocked.
Define a baseline intent. Morphic rewrites it into 5+ structurally distinct variations that preserve the original meaning but change syntax and phrasing.
Each variation is sent to your agent. We intercept the outbound JSON-RPC payload and strictly diff it against the expected schema. Every anomaly is flagged.
Precision and recall metrics produce an F1 consistency score. Drop below threshold and the CI pipeline returns HTTP 406, blocking deployment.
Captures the exact tool-call payload your agent sends and diffs it against the schema baseline. Catches hallucinated parameters, type mismatches, and missing fields.
Generate an API key, add one curl to your GitHub Action or GitLab CI config. If the consistency score drops below threshold, the pipeline fails with 406.
Configure automated test runs on a schedule. Catch silent agent drift before your customers notice. Results feed directly into the telemetry dashboard.
Track F1 scores over time across all your projects. Spot degradation trends, compare runs, and export data for incident reviews.
No SDK. No agent wrapper. Just hit the REST endpoint from any CI/CD system with your API key and baseline intent.
Forever free
Everything you need
Start catching hallucinations in 60 seconds. No credit card required.
Start Free Trial →