Both plans
- MCP (OAuth or API key), REST, and the /try playground on one meter
- Up to 2 active API keys
- Only coverage: scored is billed. Validation errors and cannot_score are free
- explain_grade and health never touch the meter
Over the cap
REST returns HTTP 402 with the Builder upgrade in the JSON body. MCP returns a score_quota_exceeded tool error that says what Builder costs. Your agent gets an error, never a made-up grade. The meter resets on the first of the month, UTC.
Connect first: MCP setup for Cursor, Claude Code, Codex, and ChatGPT.