The Pipedream alternative that writes the workflow
A Pipedream alternative for people who would rather describe a workflow than code it: the AI builds it on a canvas and verifies each step on real data.
Pipedream is a developer's automation tool: serverless workflows where each step can be a Node.js or Python function, a large library of pre-built components, and event sources that turn almost anything into a trigger. People look for a Pipedream alternative when the code-first model stops fitting. The workflow that started as three lines of JavaScript now needs a branch, a sign-off, and a retry policy, and only its author can change it. Or the team is not developers and wants the same outcomes without the code.
MadMax inverts the model: you describe the workflow, the AI writes it onto a canvas, tests every step on real data, and the result is a graph anyone on the team can read.
Where MadMax is different
You describe, it builds. One sentence becomes a plan, a graph, and the data references between steps. Max runs a verification pass and repairs what fails, stopping only where you must supply something (an account, a real sample). Hand-editing is always available. See build mode.
Verified on real data, previewed before sending. The verification run executes read-safe steps on realistic inputs, captures a real payload for webhook triggers, and previews every outbound send byte-for-byte without sending. A step passes only when its output was judged correct, so an empty object from an AI step is a failure, not a green check.
Code when you want it, not as the default. The code step is a sandboxed JavaScript transform over the payload, validated in the editor with the runtime's parser. Everything else (HTTP, branching, loops, sends, storage, AI) is a configured step whose output shape is documented and visible from any run.
Approvals as a node. The approval step pauses the run until a person decides: Slack buttons, a signed one-click email link, a YES or NO text, or the run page. The run resumes down the approved or rejected port.
AI agents with knowledge and memory. The AI agent extracts, classifies, summarizes, or runs a custom prompt with a JSON schema. It takes tools (web, browser, code, your connected apps, your other workflows), a knowledge base for retrieval, and an AI memory keyed by session.
Branching and loops on the canvas. If, switch, loop, loop until, and merge are nodes. A loop runs its body per item and fires done with the collected results, and the run page shows an iteration heatmap.
Email and text as triggers. Every workflow can mint its own inbound email address; a verified phone can text the platform number to start a run.
MCP, REST, and flat pricing. The MCP server exposes the builder's tools to Claude Desktop, Cursor, and Claude Code; the REST API ships an OpenAPI spec. Free is a personal workspace with 3 active workflows, 500 runs a month, and 100 AI credits to start. Pro is $29 per month per workspace: unlimited workflows, 2,000 runs a month, 10 parallel runs, 5 seats, 1,000 AI credits monthly, top-ups that never expire. No compute meter.
Where Pipedream is still the better choice
- Code-first by preference. If you want to write Node or Python in every step, with npm and pip packages, Pipedream is built for exactly that. MadMax's code step is a pure transform with no network access.
- Breadth of pre-built components and sources. Pipedream's component library and event sources cover a very large catalog. MadMax ships native integrations for the apps teams use most (Slack, Gmail, GitHub, Notion, Stripe, Postgres and more) plus HTTP and webhooks.
- Serverless-style execution. If your mental model is "a function that runs on an event" and you want to think about it that way, Pipedream matches it directly.
- Developer tooling. Pipedream's CLI and local component development are for developers. MadMax's developer surface is the REST API, its OpenAPI spec, and MCP.
Side by side
| MadMax | Pipedream | |
|---|---|---|
| How you build | describe it; the AI plans, builds, verifies; hand-edit any time | compose steps, write code steps, configure components |
| Testing before go-live | every step run on real data, sends previewed, samples captured | test each step against a selected event |
| Human approvals | a step: Slack, email link, text, in-app | built from suspend-and-resume patterns in code |
| AI agents | agent step with tools, knowledge bases, memory | AI provider actions and code you write |
| Branching and loops | if, switch, loop, loop until, merge as nodes | control flow in code and flow steps |
| Pricing shape | free tier; flat Pro per workspace plus AI credits | free tier; usage credits by plan |
| Integrations | native apps plus HTTP, webhooks, BYO AI models | very large component and source library |
| Self-hosting | no (enterprise gets an isolated instance) | cloud |
| API and MCP | REST API and MCP server on Pro | REST API, CLI, and MCP offerings for developers |
Moving from Pipedream
Treat each Pipedream workflow as a spec, not as code to port.
- List the workflows you still run: the trigger, what the code steps compute, where the result goes.
- Describe that to Max in plain English. Paste the gist of a code step if the logic is subtle; Max can use a code step for the part that needs it.
- Connect the accounts it asks for.
- Let verification run. For a webhook trigger, send one real event so the sample is real, then read the send previews.
- Activate, and point the source at both for a week. Compare outputs.
- Turn the Pipedream workflow off.
Frequently asked
Is there a free Pipedream alternative? MadMax has a free plan: a personal workspace, 3 active workflows, 500 runs a month, and the full integration library, no card.
Can I self-host MadMax? No. It is a managed service. Enterprise customers get a dedicated, tenant-isolated deployment on their own database and infrastructure.
Does it work with Slack and Gmail? Yes, both are native OAuth connections. GitHub, Postgres, Supabase, Linear, Notion, and Stripe are native too; any other API is reachable through the web request step.
What happens when a step fails? The run fails by default and the run page shows the failing step, its input, and the error. Per step you can choose retry (up to 5 attempts with backoff), continue out an error port, or retry then continue. See error handling.
Can I bring my own OpenAI or Anthropic key? Yes, for AI steps inside workflows. Connect Anthropic, OpenAI, Gemini, or OpenRouter and bind it on the step; usage bills to your key. The builder chat always runs on the platform model.
Related
Describe the workflow. MadMax builds it and tests it on your data.
Free to start, no card. Connect your apps, say what should happen in plain English, and watch each step run against real data before anything goes live.