Public
Audit text for orality (Havelock AI) — remote MCP server
Val Town is a collaborative website to build and scale JavaScript apps.
Deploy APIs, crons, & store data – all from the browser, and deployed in milliseconds.

Havelock Orality MCP Server

Audit text for orality — how oral/speech-like vs. literate/written it reads — powered by Havelock AI's public demo. No API key required.

The idea (via @Havelock_AI): orality is a proxy for legibility. Tell your AI:

"Iterate internally on your response and don't print it until it's above a 75 orality score."

Tool

  • score_orality{ text, include_sentences? } → 0–100 orality score, oral/literate sentence counts, and an optional per-sentence breakdown with detected oral markers (rhythm, alliteration, antithesis, rhyme, asyndeton, epistrophe, …).

Connect

Remote MCP endpoint (POST JSON-RPC 2.0):

https://havelock-mcp.val.run/

Claude Code:

claude mcp add --transport http havelock https://havelock-mcp.val.run/

Live demo: Claude scoring its own closing message

VariantScore
Max literate (one giant subordinate-clause sentence)9
Claude's default closing summary52
Tuned oral ("One tool, one job…")84
Max oral ("don't let it speak until it sings past 75")86

Notes

  • Wraps the free Hugging Face Gradio space (thestalwart-havelock-demo.hf.space): two-step submit + SSE result fetch. The first call after idle may be slow while the space wakes.
  • The Gradio API's second boolean param does not suppress the sentence breakdown — this server filters it client-side when include_sentences: false.