Voice Input for Cursor AI Editor

Cursor's AI-powered features work best with detailed context from you. When the agent asks about your project structure, preferred patterns, or what a feature should do, a quick typed "make it work" gives it almost nothing to go on. But a spoken answer naturally includes the details it needs.

"Use the existing auth middleware from the shared package. The endpoint should validate the JWT, extract the user ID, check against the team permissions table, and return a 403 if the user is not an admin. We have a similar pattern in the billing endpoints, look at those for reference." That takes 15 seconds to say. It would take a minute to type. And you'd probably skip most of it.

Because Spokenly integrates with Cursor via MCP, the agent calls it directly when it needs your input. You speak your answer, and Cursor gets the rich context it needs to write better code.

Spokenly runs a local MCP server at localhost:51089. You add it to Cursor's MCP configuration file, and Cursor's AI gets access to a voice dictation tool called ask_user_dictation.

When Cursor's agent or composer needs your input, it calls this tool. A recording prompt appears in Spokenly showing the question. You speak your answer and press Enter. The transcribed text goes back to Cursor, and the AI continues with your detailed response as context.

No bridge scripts, no terminal commands during use. Cursor connects directly to Spokenly's local server over HTTP.

Task planning: You ask Cursor to interview you before generating code. The agent walks through requirements one by one: expected inputs, error handling, UI states, data flow. You answer each question by voice in seconds and the agent assembles a full brief. Ten minutes of talking instead of an hour of writing.

Refactoring: Cursor asks "Should I move these components into a shared library or keep them colocated?" You explain which ones are reused across pages, which are one-off, and how the import paths should look. Voice captures the reasoning that a typed "shared lib" never would.

Long context: Cursor asks "What should the onboarding flow look like for team invites?" Instead of typing a couple of bullet points, you walk through each step: the email template, permissions logic, and what happens when the invite expires. The agent gets everything it needs in one go.