What it does
The detector works over precomputed dB energies and a transcript. It does not decode audio, transcribe speech, or run a trained VAD model, and its confidence values are heuristic scores rather than calibrated probabilities. It exposes four tools: analyze_turn_status, calibrate_acoustic_thresholds, predict_semantic_closure and run_benchmark_turn_detection. Each MCP process keeps its own state, and the benchmark tools run on isolated instances.
It suits work on turn-taking in a voice pipeline where energy values and a transcript are already available, and where an MCP client can call the tools. A client can read the required arguments and JSON schemas through tools/list before making calls.
The library and stdio server need Python 3.9 or newer and have no runtime dependencies. The package is installed from the GitHub release as a wheel, and the MCP client is then configured with the installed command or with the Python module entry point. The release also includes a .mcpb bundle for clients that support desktop extensions. A development check runs a deterministic example, and a separate client check exercises initialize, tools/list, tools/call and ping over stdio.
Limitations and requirements
Requires Python 3.9 or newer on PATH, or an absolute path to the installed command. The PyPI package is not yet published, so installation must come from the GitHub release. The publisher states that confidence values are heuristic, not calibrated probabilities, and that the detector does not decode audio, transcribe speech or run a trained VAD model. The deterministic example is not a protocol conformance test. Registry submissions are tracked separately, and a manifest is not proof of registry acceptance.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
genpark-voice-vad
Energy and transcript heuristics for voice turn endpoint detection.
v1.0.1As shown on registry.modelcontextprotocol.io when we last read it