MCPFlow: draw a flow, get an MCP tool

Let non-coders turn drag-and-drop flowcharts into callable MCP tools for Cursor and Claude

Problem

Writing a custom MCP tool for Cursor or Claude Desktop means setting up a TS/Python project, hand-writing schemas and wiring stdio — painful for multi-step flows like login-then-query-database. When a tool call fails it's a black box, with no visibility into which intermediate step broke.

Solution

A canvas where you drag HTTP requests, DB queries and field transforms into a pipeline, then publish it as a local MCP server that Cursor/Claude can call. Every step logs its inputs, outputs and latency, and any failed node can be replayed or re-run in isolation for debugging.

Users

Non-backend developers, PMs and analysts using Cursor/Claude Desktop for automation; indie devs tired of writing glue code for every tiny tool

MVP

Canvas with 4 node types (HTTP, DB query, JSON extract, conditional); one-click export of a Cursor-loadable MCP config; per-step execution log with single-step rerun; local-only runtime, no collaboration

Revenue

Free tier capped at 3 flows and one workspace; Pro license or subscription for unlimited flows, import/export and a template marketplace with per-use revenue share

Why now

MCP is now the de facto standard across Cursor and Claude, yet building a tool still requires coding; the MCP spec and Rust+Tauri are mature enough for one person to ship a usable canvas and runtime in 2-3 weeks