Skip to content

FAQ

Mistflow plugs into AI coding editors (Claude Code, Cursor, Codex CLI) and gives them the ability to plan, build, and deploy full web applications. You describe what you want; your AI editor builds it.

No. Mistflow itself has no AI. Your AI editor (which already has an LLM) does all the reasoning. Mistflow handles planning, building, and deploying.

No additional AI costs. You’re already paying for your AI editor (or using a free tier). Mistflow adds no token charges.

Any editor that supports the Model Context Protocol (MCP). Currently:

  • Claude Code (Anthropic)
  • Cursor (Anysphere)
  • Codex CLI (OpenAI)
  • Any future MCP-compatible tool

Yes. The code is real, generated source code. You own it completely. Push it to your own GitHub repo at any time.

By default: Next.js 15 + Drizzle ORM + Tailwind CSS + shadcn/ui with a managed database. The template provides the structure; the AI customizes everything based on your description.

On Mistflow Cloud’s global edge network. Mistflow handles deployment automatically. You get a *.mistflow.app subdomain, and can add custom domains.

Yes. Your code is a standard project. Download it and deploy anywhere. Vercel, Netlify, your own server.

Yes. You can ask your AI to preview the app locally before deploying. It starts a dev server with hot reload and can create a public tunnel for sharing.

3 projects, 3 deploys per month, 1 team member. No custom domains.

From the dashboard Settings page, or tell your AI: Upgrade my Mistflow plan.

Yes. Downgrade from the Settings page. Your projects remain accessible but you lose access to paid features.

Mistflow ships as a single MCP server (@mistflow-ai/mcp) that your editor starts as a background process. 14 discrete tools cover auth, planning, scaffolding, implementing, building, deploying, and QA. Your editor calls them over stdio/JSON-RPC — there is no separate CLI binary, no second package to keep in sync.

Long-running operations (mist_plan, mist_install, mist_build, mist_qa, mist_deploy) use fire-and-poll: the first call returns a jobId and status: "running"; subsequent calls with the jobId return current state in under a second. The MCP 60s ceiling never applies. See How It Works.

I’m getting -32001 timeout errors, how do I fix?

Section titled “I’m getting -32001 timeout errors, how do I fix?”

That error used to come from long-running operations blocking the MCP 60s per-call ceiling. MCP 1.0.0 replaced those with fire-and-poll tools, so this error should no longer appear for Mistflow tools. If you see it, upgrade: npx -y mistflow-ai install.

My existing Mistflow app doesn’t have a contracts/ directory, do I need it?

Section titled “My existing Mistflow app doesn’t have a contracts/ directory, do I need it?”

It’s optional — older projects keep working without it. The contracts/ layer is how new apps keep form validation, server actions, and the Drizzle schema in sync. New scaffolds from mist_init write it automatically.

See the contracts guide for details.

I’m on MCP 0.x and just upgraded to 1.0.0, what changed?

Section titled “I’m on MCP 0.x and just upgraded to 1.0.0, what changed?”

MCP 1.0.0 reunified the CLI back into the MCP server. If you were using @mistflow-ai/cli, every mist <cmd> command now has a direct MCP tool equivalent:

Old CLI commandNew MCP tool
mist planmist_plan
mist initmist_init
mist installmist_install (fire-and-poll)
mist implementmist_implement
mist buildmist_build (fire-and-poll)
mist debugmist_debug
mist qamist_qa (fire-and-poll)
mist mockupmist_mockup
mist deploy / mist deploy promote / mist deploy rollbackmist_deploy with action: deploy / promote / rollback
mist env <subcommand>mist_config with resource: env
mist domains <subcommand>mist_config with resource: domain

@mistflow-ai/cli is deprecated on npm; existing installs keep working but stop receiving updates. Run npx -y mistflow-ai install to get the 1.0.0 MCP server. The fire-and-poll pattern dodges the 60s timeout that motivated the original CLI split.