# Agent Message QA
> Outbound-message quality control for AI agents. Send a draft sales message + its scenario, get a reply-likelihood score (0-100), a verdict (send / send_after_minor_revision / revise_before_sending / do_not_send), named failure modes, and concrete fixes — before the message is sent.

## Endpoints
- MCP (streamable HTTP): https://api.agentmessageqa.com/mcp — tools: audit_message, report_outcome. Free calls to try; then API key.
- REST: POST https://api.agentmessageqa.com/audit-message (Authorization: Bearer <key>). OpenAPI: https://api.agentmessageqa.com/openapi.json
- Request fields: message_text (required), channel (email|linkedin|followup), target_persona, product_context, desired_outcome (book_meeting|get_reply|re_engage|qualify).

## Feedback loop
Every response carries an evaluation_id. Call report_outcome (replied true/false) 5-7 days after sending — reports earn credit and improve calibration. Only reply-status is shared, never content.

## For humans
https://agentmessageqa.com — docs, pricing, API keys.