Voxverify is a TypeScript-based tool designed to verify the correctness of Deepgram voice-AI integrations. It goes beyond checking for HTTP 200 responses and ensures that speech-to-text (STT), text-to-speech (TTS), and streaming integrations are functioning as expected.

The primary problem it solves is the common issue where voice-AI integrations return HTTP 200 status codes but still fail to deliver the correct results. This can happen due to empty transcripts, incorrect transcripts, JSON error envelopes, or other issues that are not caught by a simple status code check.

The tool is used by developers and AI coding agents who need to ensure their voice-AI integrations are working correctly. It can be used as a CLI tool or as an MCP server, allowing AI agents to verify their own work honestly.

Voxverify works by performing various checks on the integrations. For STT, it transcribes known-answer audio and asserts the transcript matches the expected result. For TTS, it checks that real audio is returned. For streaming, it ensures that a correct final transcript or audio stream is delivered. The tool also includes an offline config check to validate models and parameters without requiring an API key.

Notable patterns include the use of a word error rate (WER) threshold for STT transcript comparison, the inclusion of a mock transport for offline testing, and the use of the Model Context Protocol (MCP) for agent integration. The tool is designed to be used in CI pipelines and provides detailed reports on the verification results.