Pipedream

Section 2 — Is this for you?

Pipedream is the automation tool you pick when you can write code and don't want to pretend otherwise. Every workflow is a chain of steps where "drop in a Node.js or Python snippet" is a first-class option — not a last-resort escape hatch buried three menus deep. If your team has a developer, Pipedream stops being a low-code tool and starts being a serverless runtime with 3,000+ pre-wired integrations on top.

It's also where the "put AI agents on top of real business APIs" story is furthest along. Pipedream's MCP Server exposes its entire integration catalog — ~10,000+ tools across ~3,000 apps — to MCP-aware AI clients with managed OAuth. If you're building agents in 2026, that's not a nice-to-have.

Heads up: Workday announced a definitive agreement to acquire Pipedream on November 19, 2025. Expected to close by end of January 2026. The product is running business-as-usual today, but Workday's direction will shape the roadmap. Worth factoring into long-horizon bets.

Pipedream is the right pick when:

• You have at least one developer on the team and want code-first workflow building
• You're embedding integrations in a product you sell (Pipedream Connect is built for this)
• You're building AI agents and want production-grade MCP with managed OAuth for 3,000+ apps
• You need custom logic that doesn't fit in a pre-built action — Node.js, Python, Go, or Bash run inline
• You want per-compute-second billing instead of per-task — cheaper if your flows are fast

Pipedream might work, but watch for:

• Your team is non-technical — the Node.js/Python step is the main reason to pick Pipedream, and the UI assumes you want to look at code
• You need the polished, white-label embedded experience — Pipedream Connect still shows "Pipedream" branding during the auth flow
• Your workflows run >12 minutes — that's the hard ceiling; plan around it or move to a different runtime
• You need EU data residency with a signed DPA — Pipedream runs primarily in US AWS regions as of early 2026

Look elsewhere when:

• You want non-technical teammates to build their own automations end-to-end — Zapier is simpler
• You want a visual 2D canvas with routers and aggregators — Make is the canvas tool
• You want to self-host the whole platform for free — n8n is the only real choice
• You're deep in the Microsoft 365 stack — Power Automate plugs into it natively

Section 3 — The real story

Pipedream started life as RequestBin — the open-source HTTP request inspector a lot of developers used for years to debug webhooks. Pipedream the company launched the public SaaS version in October 2019, with the original service quietly running from December 2018.

The initial pitch was narrow: give developers a webhook URL, inspect what comes in, do something with it. Then Tod Sacerdoti and the team kept adding — code steps, event sources, scheduled workflows, pre-built integrations, Connect for embedding. By 2024 it had become a full-blown workflow platform targeted at developers, with an active free tier, thousands of components, and a vocal community.

Pipedream's identity is developer-first. You can build a workflow without writing code — drop in pre-built actions, configure them in the UI — but the moment you need custom logic, you write it. There's no "upgrade to premium to use code." The free tier ships with Node.js, Python, Go, and Bash steps.

November 19, 2025: Workday signed a definitive agreement to acquire Pipedream. The deal is expected to close by end of January 2026 (Workday's fiscal year-end). Workday's pitch is that Pipedream's 3,000+ connectors and MCP-native architecture let Workday's AI agents act across Asana, HubSpot, Jira, Slack, Recurly, and effectively any app with an API — not just inside Workday itself.

For buyers today, the acquisition means three things in practice:

• Pipedream's runtime and product keep shipping normally through close and for the foreseeable future
• Enterprise buyers will increasingly hear about Workday integration as a differentiator
• Independent companies that compete with Workday (HR tech, finance tech) should pay attention — some may want a non-Workday alternative by late 2026

The Connect pivot

Throughout 2024 and 2025 Pipedream invested heavily in Pipedream Connect — an embedded product that lets you drop a "connect your apps" widget into your own SaaS product and have your customers auth their accounts on your behalf. Connect, plus Pipedream's MCP Server, is what made Workday want to buy them. The core workflow product funded it; Connect became the strategic lead.

Section 4 — How it works

Workflows are step lists you can drop code into

A Pipedream workflow is a trigger plus an ordered list of steps. Each step is either a pre-built action (from Pipedream's catalog of 3,000+ integrated apps) or a custom code step in Node.js, Python, Go, or Bash.

The builder looks a lot like n8n or Power Automate's vertical step list, but with one big difference: clicking "new step → run custom code" drops you into a full code editor with access to the event payload, previous steps' outputs, connected accounts, and npm/PyPI packages.

Triggers and event sources

• HTTP trigger — gives you a unique webhook URL. Any HTTP method, any payload. This is Pipedream's original use case and still its most popular trigger.
• Schedule trigger — cron syntax, runs the workflow on a timer. Default timeout is 60 seconds.
• Email trigger — unique inbox address; emails sent there run the workflow with the parsed message as the event.
• App-based triggers — "new row in Google Sheets," "new message in Slack," and so on. Built on Pipedream's event sources layer — a second-level primitive where community or first-party code polls/streams events from external apps.
• Custom event sources — advanced: write your own source in Node.js, deploy it, and have it feed any number of workflows.

Code steps

This is the headline feature. In any step you can:

• Run Node.js v20 with access to most of npm's 400,000+ packages
• Run Python with the Pipedream SDK and most of PyPI
• Run Go or Bash for edge cases
• Import connected accounts so OAuth tokens are injected at runtime

The code gets the incoming event and all prior steps' outputs as inputs, and whatever you return or assign to the step output flows to the next step. This is as close to "serverless function with integrations bolted on" as any workflow platform gets.

Timeouts, memory, and compute

• Default timeout: 30 seconds for HTTP / email triggers, 60 seconds for cron triggers
• Max timeout: 12 minutes (plan-dependent ceiling; some plans go higher)
• Default memory: 256 MB per workflow
• Max memory: 10 GB (plan-dependent)

Increasing memory proportionally increases CPU, so a CPU-bound workflow can finish faster if you pay for more memory — and Pipedream's pricing accounts for that by scaling credit consumption with memory.

Data Stores for persistent state

Pipedream has a built-in key-value store called Data Stores — Brotli-compressed, JSON-serializable values you can read and write from any workflow in a project. This is how you track state across runs (idempotency keys, rolling aggregates, last-synced timestamps) without standing up a real database.

For heavier state you bring your own database — Pipedream has connectors for Postgres, MySQL, Supabase, MongoDB, and most of the usual suspects.

Branching and loops

• If steps give you conditional branching.
• Loops are just code — write a for loop in a Node.js step, call an action inside it. This is unusual compared to the explicit "Apply to each" / "Iterator" UI primitives in Power Automate or Make.
• Retry and delay are configurable per step.

This is the strength and weakness of the platform: code gives you unlimited flexibility, and also means that when something breaks, you're debugging like a developer, not clicking through a visual trace.

Section 5 — Where it shines

Code is a first-class citizen

No other workflow platform on this list treats code this well. Zapier has Code steps (JS and Python) but they're bolted on. Make has JSON parsers and text aggregators but no real code. n8n has a great Code node — Pipedream matches it and ships it in the free tier.

If your team thinks in functions and event payloads, Pipedream will feel like home in about ten minutes. Reach for npm, import a library, return a value, move on. The cognitive overhead is close to zero.

MCP leadership for AI agents

Pipedream has arguably the most mature Model Context Protocol story in 2026. The Pipedream MCP Server exposes 10,000+ tools across 3,000+ apps, handles OAuth token management, and is already referenced in Claude, ChatGPT, Cursor, and other MCP-aware clients' ecosystems.

For teams building agents, "I need my agent to create a HubSpot contact and post to Slack" goes from a week of glue code to a few lines of MCP client setup. That's the strategic reason Workday bought the company.

Pipedream Connect for embedded integrations

If you're shipping a SaaS product and your customers ask for "integration with Salesforce / HubSpot / Slack / Google Drive," you have three realistic options: build them yourself (expensive), buy a unified API like Merge (covers categories), or embed Pipedream Connect (covers 3,000+ apps individually, with managed auth).

Connect handles OAuth flows, stores encrypted tokens for your end users, exposes a JavaScript SDK and a REST API, and lets you drop a "connect your apps" widget in your app. Your end users authenticate once; your code calls Pipedream-hosted actions against their accounts.

The Nango team fairly points out Connect isn't fully white-label and observability is thinner than purpose-built alternatives. Balance that against Connect's depth — nothing else has this many pre-built connectors with a free tier.

Per-compute-second billing, not per-task

Pipedream charges one credit per 30 seconds of compute at 256 MB. A workflow that finishes in 200 ms consumes 1 credit. A workflow that runs for 25 seconds consumes 1 credit. A workflow that doubles its memory to 512 MB consumes 2 credits for the same run time.

For fast workflows — a webhook, three API calls, a Slack notification — this is dramatically cheaper than Zapier's per-task or Make's per-operation model. A flow that would be 5 Zapier tasks ($$) is often 1 Pipedream credit (fraction of a cent).

The flip side: long-running workflows (data backfills, LLM calls with retries, slow external APIs) burn credits fast. Budget accordingly and watch your usage.

Generous free tier for developers

Pipedream's free tier is the most useful of any platform on this list. 100 credits per day, 3 workflows, unlimited connected accounts for development. Most developers can build real automations on the free tier for months before they hit limits. That's by design — the go-to-market story is "developers adopt it for side projects, their companies pay for it later."

Github-style project model

Workflows live in projects with their own Git-backed source of truth. You can export a project to a repo, version it, code-review changes, and deploy via Pipedream's CLI. For teams that treat automations as production code (which Pipedream's user base mostly does), this is how ALM should work.

Section 6 — What breaks most often

Cold starts on the free and lower tiers

The first request to a workflow after ~5 minutes of inactivity triggers a cold start — Pipedream spins up a fresh worker to handle the event, which takes a few seconds.

For interactive workflows (Slack slash commands, user-facing webhooks) this is the single most common complaint. The fix is dedicated workers — keep a worker warm at all times so there's no cold-start penalty — which costs credits and is only available on higher-tier plans. New dedicated workers also take 5–10 minutes to fully configure, so the warm-up isn't instant when you first provision.

The 12-minute ceiling

Cloud workflows max out at 12 minutes per execution. If your job takes longer — big backfill, slow LLM chain, external API pagination over thousands of rows — you'll hit the wall.

The workaround: chain multiple workflows together, one triggering the next with a cursor or continuation token. It works, but it's more plumbing than a platform with no hard timeout.

Timeout surprises on the default

The default timeout is 30 seconds for HTTP / email, 60 seconds for cron. Most developers don't change this on workflow #1, then discover six months later that an action started hitting a slower API and the flow silently fails.

Always raise the timeout explicitly for any workflow that does more than 1–2 API calls, and add retries on transient errors.

Credit math is harder to predict than per-task

Per-compute-second billing is great when your workflows are fast and cheap to reason about when you've been running them for a month. It's harder to predict up front than Zapier's per-task model. "How many credits will this workflow use?" is a real question and the honest answer is "depends on how long it runs, and memory settings."

Watch the usage dashboard in the first few weeks of anything you deploy to production.

Connect is not fully white-label

If you embed Pipedream Connect and your end users go through the auth flow, they'll see Pipedream's name in the OAuth consent screen and in some parts of the UI. For many SaaS use cases this is fine. For regulated-industry or enterprise buyers who expect the integration to look like part of your product, this is a blocker.

Paragon and Workato are more white-label; Nango is more developer-oriented. Pipedream is the "broadest catalog with managed auth" option, not the "invisible branding" option.

Observability is thinner than enterprise platforms

Compared to Datadog-grade observability, Pipedream's run logs, error tracking, and request tracing are lightweight. You get the basics — run history, step outputs, error messages — but deep tracing and OpenTelemetry export aren't first-class in 2026.

If you're running Pipedream workflows in a regulated prod stack, plan to send important events to your own observability pipeline (Sentry, Datadog, Honeycomb) from inside the workflow.

EU data residency is limited

Pipedream's production infrastructure runs primarily in US AWS regions. Enterprise customers can negotiate US data residency contracts; EU-only residency is not a standard offering. For EU-headquartered or EU-regulated buyers, this is often the first question compliance asks and the first reason a deal stalls.

Section 8 — Getting started reality

The fastest path: a webhook in two minutes

Go to pipedream.com, sign up with Google or GitHub (the free tier is immediate), and click New Workflow. Pick HTTP/Webhook as the trigger. You get a unique URL. Send anything to it with curl — Pipedream shows you the incoming event.

Add a second step. Choose "Run Node.js code" if you're a JS developer, or pick a pre-built action like "Send a message to Slack." Connect the Slack account when prompted. Deploy.

That's it. The workflow is live. Send another request to the webhook URL and you'll see it run end-to-end in a few seconds.

Your first real workflow

Most developers' first real Pipedream workflow is a glue job — something like "when Stripe fires a webhook, look up the customer in HubSpot, post a summary to Slack, add a row to a spreadsheet." That's a five-step workflow: Stripe trigger, HubSpot action, a custom Node.js step to format the message, Slack action, Google Sheets action.

Build it in the UI, test each step with sample events, deploy. Total build time is usually 30–60 minutes. Total cost at production load is usually under a dollar a month.

Using Connect

For embedded integrations, read the Connect overview. The quickstart involves:

1. Creating a Connect project in Pipedream
2. Adding an OAuth client for each app you want to support (bring-your-own or use Pipedream's)
3. Using the JavaScript SDK to render the auth UI in your app
4. Calling Pipedream actions server-side with the access token

It's well-documented but has more moving parts than "set up a single workflow." Expect a real day of work to ship a production Connect integration.

Where people get stuck

• Cold starts on the free tier. First request after idle is always slow. Expected; document it for users; pay for dedicated workers on anything interactive.
• Default timeout is too short. Raise it on any workflow with external API calls.
• Python props not in UI. Props (input fields in the workflow builder) only work for Node.js steps. Python steps have to read inputs differently.
• Connect branding surprise. Realize before launch that your end users will see "Pipedream" during OAuth. Test the full flow with a real customer before you ship.
• Credit usage spikes. A buggy loop that retries a slow API can burn thousands of credits overnight. Add alerts on daily credit usage.

Free learning resources

• Official Pipedream docs — clear, code-heavy, good examples
• Pipedream blog — frequent releases and deep-dives, especially on Connect and MCP
• Community forum — active, with direct Pipedream engineers answering
• Pipedream GitHub — the entire integrations library is open source; great for understanding how actions work

Section 9 — Migration translations

From Zapier

• Zap → Workflow. Same concept. Pipedream's editor feels more like a code editor than Zapier's form-driven UI.
• Task → Credit. Radically different model. Zapier bills per successful action; Pipedream bills per 30 seconds of compute. Fast workflows are much cheaper on Pipedream; slow ones (long LLM calls) can be more expensive.
• Paths → If steps or code branching. Zapier Paths become Pipedream's If steps, or just an if statement inside a code block.
• Code by Zapier → Code steps (native). First-class feature on Pipedream. No paywall. Full Node.js, Python, Go, and Bash.
• Zapier Tables → Data Stores. Pipedream's built-in key-value store covers most Tables use cases; for anything heavier, connect a real database.
• Zapier Interfaces → Not a direct equivalent. Pipedream has no form/interface builder. Use a separate tool like Retool, or render forms from your own app.

From Make

• Scenario → Workflow. Same concept, very different UI. The 2D canvas becomes a vertical step list. Make users often miss the visual overview.
• Router → If / code branching. No explicit router UI; use If steps or branch in code.
• Iterator → for loop in code. In Pipedream you loop in Node.js. Faster to write for developers; less discoverable for non-developers.
• Aggregator → Code step that collects results. Same pattern in code: accumulate into an array, pass to the next step.
• Operation → Credit (per 30s compute). Different counting model. Usually cheaper on Pipedream if your workflows are fast.
• Data Store → Data Store (same name, same idea). The mental model transfers directly.

From n8n

• Workflow → Workflow. Same concept. Pipedream is cloud-only; n8n can self-host. If self-host is why you chose n8n, Pipedream is not the move.
• Code node → Code step. Nearly identical capability. Both support Node.js and Python. Pipedream has a broader npm install surface on the free tier.
• Nodes → Actions. Same idea. Pipedream's catalog is bigger (3,000+ vs n8n's ~400 official).
• MCP Server Trigger → Pipedream MCP Server. Both support MCP; Pipedream's managed service is more mature in production. n8n's is better for self-hosted agents.
• Custom nodes → Custom components. Pipedream's components are open-source in the Pipedream GitHub repo. You can submit PRs or build private components for your team.

From Power Automate

• Cloud flow → Workflow. Same concept. The vertical step list is similar.
• Connector → App / action. Terminology differs; capability is similar. Pipedream has no "premium vs standard" split — all 3,000+ apps are available at every tier.
• Condition / Switch → If / code branching. Power Automate's Condition maps to Pipedream's If. Switch becomes a code block or chained Ifs.
• Apply to each → for loop in code. Pipedream auto-wraps nothing; you explicitly loop when you want to.
• Approvals → Not a direct equivalent. Power Automate's native Approvals (Teams / Outlook) have no one-click equivalent in Pipedream. Build with Slack-based approvals or a custom UI.
• Desktop flows → Not available. Pipedream does not do RPA. If you need to automate a legacy desktop app, stay on Power Automate for that part.
• API request quota → Credits. Different counting model. Pipedream's per-compute-second billing is usually more predictable than Power Automate's daily per-user API quota.

Section 10 — AI & MCP readiness

MCP — Pipedream is in front

The Pipedream MCP Server is one of the most-deployed MCP servers in production as of early 2026. It exposes 10,000+ tools across 3,000+ apps — the entire Pipedream integration catalog — to any MCP-aware client (Claude, ChatGPT, Cursor, custom agents).

Managed features that matter:

• OAuth handling — Pipedream manages the full auth lifecycle for each app, so your agent doesn't need to store refresh tokens
• Per-user credentials — each end user's connected accounts are isolated; agents act on behalf of the authorizing user, not a shared service account
• Encrypted credential storage — tokens are encrypted at rest; requests proxy through Pipedream, never exposing credentials to the LLM
• Usage and logging — basic observability for which tools the agent called, when, and with what result

For teams building AI agents that need to take real action across SaaS — "book a meeting, create a Jira ticket, send a Slack message, update the CRM" — Pipedream's MCP server is the path of least resistance.

MCP Chat and the open-source angle

Pipedream publishes MCP Chat, an open-source reference agent built on the AI SDK plus Pipedream's MCP. Companies fork it to bootstrap their own agent products — it saves weeks of plumbing.

LLM tokens and AI Builder

Pipedream paid plans include an AI token quota (20M/month on Basic, 50M/month on Advanced). The tokens are consumed by Pipedream's AI Studio features — built-in AI actions that call LLMs with your prompt templates and typed inputs/outputs, no API key management required.

You can also plug your own OpenAI, Anthropic, Google, or other provider keys into a Pipedream workflow as a connected account and call them from any code step. Many Pipedream workflows are, in practice, LLM-first — "send this incoming email to Claude, extract intent, route it."

The Workday angle

With the Workday acquisition closing, expect Pipedream's AI story to increasingly emphasize agents that act across Workday plus external systems. The product is still independent in the market today, but the 2026 roadmap will tilt toward "AI agents that bridge Workday and the rest of your stack."

The summary

For developer-led teams that want to build AI agents with real action capability, Pipedream's MCP story is as good as anything on the market. The only platforms in the same league are the underlying LLM providers themselves (via their own tool-use APIs) and purpose-built MCP platforms like Composio or Klavis.

Where it's thinner: UI for non-technical agent builders. If your business team wants to build an agent without a developer in the loop, Copilot Studio or Zapier's AI features are friendlier.

Section 11 — Compliance at a glance

The facts

• SOC 2 Type II (Security): Yes — report available to enterprise customers
• SOC 2 Type I (Confidentiality + Availability): Yes (Type II coverage expanding in 2025+)
• ISO 27001: Available on Enterprise
• GDPR: Yes — standard DPA available
• HIPAA / BAA: Yes, on Enterprise plans only — request via Pipedream support
• Data residency: US primary (AWS). EU-only residency not a standard offering.
• FedRAMP: No
• SSO (SAML / OIDC): Yes — available on Business and Enterprise plans
• RBAC: Yes — project and workspace-level roles
• Audit logs: Yes — on Enterprise
• Trains AI on your data: No — Pipedream does not train models on customer workflow data
• Self-hostable: No — cloud-only. Dedicated infrastructure is negotiable on Enterprise; full self-host is not.
• SLA: 99.95% uptime on Enterprise

The fine print

Pipedream's compliance footprint is middle-of-the-pack: solid enough for most SaaS and developer use cases, lighter than Power Automate or Workato for regulated-industry enterprise buys.

The HIPAA BAA is the one that catches teams off guard. Unlike Power Automate (where the BAA ships automatically with M365) or Zapier (where it's a standard Enterprise add-on), Pipedream's BAA is Enterprise-tier only and requires a support request to activate. Covered entities planning to pass PHI through Pipedream need this conversation up front, not at launch.

EU data residency is the second common blocker. Pipedream's production runs primarily in US AWS regions. For EU-headquartered or EU-regulated buyers — especially in financial services and healthcare — this is often where the deal stalls. Workday's EU infrastructure may change this story post-acquisition, but as of early 2026 Pipedream is not a first-choice option for EU-only residency.

SOC 2 Type II on Security is where Pipedream is solid. The report is available under NDA and covers the standard controls most buyers care about. Confidentiality and Availability Type II is rolling in over 2025-2026.

For team management — SSO, RBAC, audit logs — everything lives on the Business and Enterprise tiers. The Free, Basic, and Advanced plans are developer-first and don't include enterprise governance. If you're spinning up Pipedream for a team of more than a handful of developers, budget for Business at minimum.

The one structural tradeoff: Pipedream is not self-hostable. Cloud flows only run in Pipedream's cloud. For compliance regimes that require full on-prem or VPC-level isolation, the answer is either n8n (self-host) or a dedicated-infrastructure negotiation with Pipedream Enterprise. "Fully in our VPC, we hold the keys" is not a standard Pipedream offering — it's a custom conversation.