Talkform MCP server

Talkform

Turn any form into a live audio interview.

Talkform is a config-driven audio intake utility. Products define the fields they need, Talkform asks for them conversationally, binds structured values directly into the form, and exports JSON-ready session results.

Workspace

• apps/web: hosted app for talkform.ai
• packages/core: config schema, realtime prompt/tool generation, session/result helpers
• packages/react: embeddable React widget
• packages/http: process-local transient session and HTTP helpers
• packages/cli: agent-friendly CLI
• packages/mcp: local MCP schema, template, and config-validation server
• docs: markdown documentation rendered in the web app

Core commands

pnpm install
pnpm dev
pnpm build
pnpm typecheck

Environment

• OPENAIAPIKEY
• OPENAIREALTIMEMODEL default gpt-realtime-2.1
• OPENAIREALTIMEVOICE default marin
• TALKFORMENABLEIMPORTAIREFINEMENT optional; only true enables paid AI copy refinement after deterministic import
• NEXTPUBLICAUDIOFORM_VENDOR
• NEXTPUBLICAUDIOFORMVENDORURL

Transient session APIs and public Realtime client-secret issuance are disabled in hosted production by default. A production deployment needs a durable session store, a distributed rate limiter, and server authentication before those endpoints should be enabled. The current in-memory store and quotas are development/reference implementations; they do not provide cross-instance durability or enforcement.

Machine clients use two matching environment variables:

• TALKFORMAPITOKEN: server-only bearer token accepted by the web API
• AUDIOFORMAPITOKEN: client-side token read by the Talkform CLI

For a controlled deployment, TALKFORMENABLEINMEMORYSESSIONS=true and TALKFORMENABLEPUBLIC_REALTIME=true opt into the reference endpoints. Do not use these flags to represent the current process-local implementations as production-ready infrastructure.