@mistflow-ai/mcp

The Mistflow MCP server — an app-building tool provider for AI coding editors (Claude Code, Cursor, Codex CLI, VS Code Copilot).

Most users should not install this package directly. Use the mistflow-ai installer instead: npx -y mistflow-ai install. That command auto-detects your editor and wires everything up.

What this is

A Model Context Protocol (MCP) server that exposes Mistflow's app-building capabilities as tools to your AI coding editor. Your editor's AI does the reasoning; this server provides the short, structured control-plane flows, while the companion CLI handles planning, scaffolding, building, seeding, and deployment.

The server never calls an LLM itself. It's a tool orchestrator, not a brain.

Tools exposed

| Tool | Purpose | |------|---------| | mist_setup | Device-code auth flow — log in via browser, credentials stored in ~/.mistflow/credentials.json | | mist_project | Read/update project state so the host AI can reason about an existing Mistflow app | | mist_browser | Browser automation for testing the live app | | mist_help | Return the Mistflow CLI catalog so the host AI knows which mist commands to run next |

Manual install (if the installer doesn't work for you)

Run:

claude mcp add mistflow --scope user -- npx -y @mistflow-ai/mcp@latest

Edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "mistflow": {
      "command": "npx",
      "args": ["-y", "@mistflow-ai/mcp@latest"],
      "transport": "stdio"
    }
  }
}

Edit ~/.codex/config.toml:

[mcp_servers.mistflow]
command = "npx"
args = ["-y", "@mistflow-ai/mcp"]

Edit .vscode/mcp.json in your workspace:

{
  "servers": {
    "mistflow": {
      "command": "npx",
      "args": ["-y", "@mistflow-ai/mcp@latest"],
      "type": "stdio"
    }
  }
}

Restart your editor after editing. Then say "Log me in to Mistflow" to authenticate.

Configuration

The server reads from environment variables and ~/.mistflow/credentials.json.

| Variable | Purpose | |----------|---------| | MISTFLOW_API_KEY | Skip browser auth entirely — useful for CI. Prefix: mist_. Create a key at app.mistflow.ai/mcp-keys. | | MISTFLOW_API_URL | Override the backend URL (default https://api.mistflow.ai) |

Command-line flags:

npx @mistflow-ai/mcp --api-url http://localhost:9100  # local dev

Security notes

• Credentials are stored in ~/.mistflow/credentials.json (readable only by the current user).
• API keys prefixed with mist_ are not logged, not echoed, and redacted from any error output.
• The server never writes to stdout — all logging goes to stderr so it can't corrupt the MCP JSON-RPC protocol.

Links

• Website: mistflow.ai
• Docs: mistflow.ai/docs
• Installer: mistflow-ai

License

Elastic License 2.0. Source-available, with one restriction: you may not offer @mistflow-ai/mcp as a hosted or managed service that competes with Mistflow. Everything else — using it inside your own apps, modifying it, embedding it in products you sell — is allowed. Paying Mistflow customers are unaffected.

Copyright (c) 2026 Mistflow.ai