Skip to content

MCP Tools Overview

Mistflow exposes 13 tools to your AI editor via MCP. Your editor discovers these automatically — you never call them directly.

Your AI editor sends JSON-RPC tool calls to the Mistflow MCP server running locally. The server validates inputs, calls the Mistflow API when needed, and returns structured results. The editor uses these results to continue reasoning.

Editor → tool call → MCP Server (local) → Mistflow API (cloud) → result → Editor

plan-app

Takes a natural-language description and returns a structured build plan with features, tech stack, and implementation steps. Reference →

init-project

Scaffolds a new project from a template using the plan. Creates the file structure, installs dependencies, initializes git. Reference →

implement-step

Implements a single step from the plan. The AI writes the code; Mistflow validates it builds and tracks provenance. Reference →

deploy

Builds and deploys the project to Cloudflare Pages. Returns the live URL. Reference →

preview

Starts a local dev server with hot reload and optionally a Cloudflare tunnel for sharing. Reference →

debug-build

Runs the build, captures errors, and returns human-readable diagnostics with fix suggestions. Reference →

undo-last-change

Reverts the last implement-step by resetting to the previous git commit. Reference →

get-project-state

Returns the current project state: plan, completed steps, deploy status, and config. Reference →

update-project-state

Updates project metadata or plan state (e.g., marking a step as done). Reference →

get-provenance

Retrieves the provenance trail for a project — which prompts built which features. Reference →

connect-github

Connects a GitHub repository to the project for auto-sync on deploy. Reference →

manage-domain

Adds, verifies, or removes custom domains for a project. Reference →

The setup tool handles first-time authentication — it opens the browser for login and stores the API key locally.