Fabriq Developer vs Pipedream: The Free Tier, the Credits and the Tool Surface

TL;DR

Pipedream's free plan does not run in production, and that single fact decides most of this comparison. The pricing page marks the free tier "Development only", and the Connect docs cap the development environment at 10 external user IDs, with the caveat that you "must be signed in to pipedream.com in the same browser" while connecting. Real customers need the Startup plan at $99 a month.

The second fact is that Pipedream prices an agent's work in compute. A tool call through MCP costs 1 credit per 30 seconds, Startup includes 10,000 credits, and additional credits are $0.012 each. That's a workflow runner's meter attached to a loop that may call ten tools to answer one question.

The usual framing of Pipedream as "the workflow platform with a huge connector catalogue" is now out of date in a way that matters. Pipedream announced that Workflows and String shut down on 31 March 2027, with all Workflows data, including connected-account credentials, deleted by 30 April 2027. Connect is explicitly unaffected and is where the company says its focus is going.

None of this makes Pipedream the wrong choice. 3,000+ APIs against our 509 is not close, and its end-user connect flow reaches all of them where ours reaches thirteen. Pick on volume shape: how many users, how many tool calls each, and whether you need the long tail today.

Overview

Our other pages on this market, 10 Best AI Agent Integration Platforms and Composio alternatives, rank platforms on identity, per-user authority and audit. That's the right axis for a security review and the wrong one for a seed-stage team picking something on a Friday. This page is the Friday version: free tier, catalogue, time to first tool call, OAuth paperwork, and who holds the token.

Pipedream and Agentic Fabriq arrive at the same job from different histories. Pipedream spent years as a workflow runner with an enormous component registry, then built Connect on top of it: a managed-auth layer with SDKs and a remote MCP server. Fabriq Developer never had the workflow half. It's a gateway that holds your users' credentials and serves tools to an agent over one MCP endpoint.

That history still shows in the billing unit, in how the tool surface is addressed, and in what the free tier is for. Those are the three things we'd check before writing any code, so that's the order below. Every number here comes from a page we opened in October 2026, and where a figure isn't published we say so rather than guess.

What You're Buying Now

Start with the thing that would have been wrong to write from memory. Pipedream's docs carry a notice titled "Workflows and String Are Shutting Down on March 31, 2027". The reasoning is stated directly: "We've decided we're focusing on being the integration layer companies rely on to connect AI to the outside world, and Workflows falls outside that focus."

The operational details matter if you were planning to use workflows as glue. Workflows are in maintenance mode now, with no new signups or subscriptions and no new features. Existing ones keep running until the shutdown date. All Workflows data is permanently deleted by 30 April 2027, and the notice is explicit that this includes "the credentials and OAuth tokens for your connected accounts". There's an export tool that downloads workflow structure as JSON, one project at a time.

Connect is carved out in the first line of the same notice: "If you're using our Connect API, MCP Servers, or API Proxy, nothing changes for you." So this isn't a company in trouble, it's a company that picked a lane, and the lane it picked is the one this comparison is about.

Two practical consequences. First, if your plan involved an agent calling a tool that triggers a workflow that does the real work, that architecture has about five months left as something you can still build on. Second, the credit model you're about to be billed under was designed for workflow execution, and it hasn't changed shape now that the thing being executed is a tool call from a model. That's the next two sections.

The Free Tier Is a Development Environment

Pipedream's free plan gives you a lot and withholds one thing. The pricing page lists all build features, 3,000+ connectors, 10,000+ prebuilt tools, unlimited team members and community support, then marks the production environment and usage credits "Development only".

The Connect docs put numbers on what development means. The environment is "limited to 10 unique external user IDs", and if you exceed that you have to delete some before adding more. When connecting an account in development mode you "must be signed in to pipedream.com in the same browser" where you're connecting. And the docs state the intent plainly: "Development mode is intended for your own accounts during testing and development, not for your real end users", with a warning that the environment "is not intended for production use with your customers."

Read that against a seed-stage reality. You cannot put a development-environment Connect integration in front of ten design partners, because they'd each need a pipedream.com session in the same browser. The free plan is for you, on your laptop, with your own Gmail. It is a good free plan for that, and it is not a free plan you can ship.

Production means Startup, at $99 a month billed monthly or $90 a month on annual billing, which includes production access, 100 users and 10,000 usage credits, with "then $2 per user" beyond the hundred. Business is custom and annual-only, adding SSO, SCIM, HIPAA, dedicated static IPs and SLAs.

Fabriq Developer's free plan is shaped the other way. It's free forever with 20,000 API calls a month, unlimited agents, per-agent tool permissions, request logs, live sessions, the Python SDK, the afctl CLI and the MCP endpoint, and it runs in production with real customers on day one. There's no separate development environment and no per-user charge in the published plans. The paid steps are $50 a month for 200,000 calls and $300 a month for 2,000,000.

One honest asymmetry: triggers, the provider-event subscriptions that call your webhook, sit on our Launch plan rather than the free one. If you need to react to a provider event before you're paying anyone, Pipedream's free development environment will let you prototype that and ours won't.

When an Agent Loops

Here's the part we'd model in a spreadsheet before committing, because it's where a workflow runner's pricing meets an agent's behaviour.

Pipedream's unit is the credit, and the definition on the pricing page covers exactly the operations an agent performs: "Running an action, calling a tool through MCP, sending a request through the Connect proxy, and emitting an event from a deployed trigger each use 1 credit per 30 seconds of compute". The docs add the useful other half, which is what's free: listing apps, actions and triggers, listing accounts, configuring actions and triggers, and other management operations consume no credits. So discovery is free and execution is metered.

A tool call that returns in two seconds and a tool call that returns in twenty-nine both cost one credit. Startup includes 10,000 of them. Beyond that, additional credits are $0.012 each.

Now put an agent on it. A single user request that lists messages, reads three of them, searches a second system and writes one record is six tool calls. Retries and a model that re-reads before writing push that higher. At six calls per request, Startup's included credits cover roughly 1,666 user requests a month across your entire customer base. A product doing 300,000 tool calls a month pays $99 plus 290,000 additional credits at $0.012, which is $3,480. That arithmetic is ours, from Pipedream's published rates, and it assumes every call finishes inside 30 seconds. Slower calls cost more.

Fabriq's unit is coarser and we'd rather be upfront about that than win a comparison on a technicality. A call is any request your agent makes through Fabriq, which includes authenticating and retrieving credentials, not only the tool call. Against that, 2,000,000 calls is the $300 tier. The duration of a call doesn't change what it costs, which is the structural difference: we meter requests, Pipedream meters compute.

Figure 1 — Pipedream meters compute time, Fabriq meters requests. An agent that makes many short calls is the worst case for one model and the normal case for the other.

There's a second line on the invoice worth planning for. Pipedream bills external users separately from credits, where an external user is "each distinct user of your app who connects at least one account". Startup includes 100 and charges $2 per user after that. Five thousand connected users is 4,900 chargeable users, or $9,800 a month before a single credit. Fabriq's published plans have no per-user line, which is the single biggest cost difference between the two at consumer-ish scale.

One App at a Time

The way Pipedream's tools reach a model is the second place its origins show, and it's the detail we'd most want to know before building.

Pipedream's remote MCP server lives at https://remote.mcp.pipedream.net/v3 and speaks both SSE and streamable HTTP with no configuration. Four headers are required on every request: x-pd-project-id, x-pd-environment, x-pd-external-user-id and x-pd-app-slug. That last one is the interesting one. The docs describe each MCP server as corresponding to "an app integration (like Notion, Gmail, or Slack)" with "its own specific set of tools", so the app is part of the address, not something the agent discovers inside a single connection.

In practice that means your code decides which app's tools the model can see before the model sees anything. If an agent needs Gmail and Notion and Linear, that's three servers, or one server re-addressed per turn. Pipedream gives you an app discovery API to pick the slug programmatically, and apps.list with a query string is a perfectly good router. But the routing is your job, and an agent that doesn't know which app it needs until it has read the user's message has to be built around that.

The compensating design is genuinely clever. If a user has no connected account for a tool the agent tries, Pipedream's MCP server returns a Connect Link URL in the tool-call response, with no extra implementation required. The agent can hand that link straight to the user mid-conversation. We don't have an equivalent in the tool response and we think it's the nicest thing in Pipedream Connect.

Fabriq's surface is addressed per user instead of per app. for_user(request) mints a short-lived user-scoped token, 900 seconds by default with a one-hour ceiling, and the session serves every tool that user granted across every provider, with the effective list being the intersection of the user's scopes and the agent's. One endpoint, one tools/list, no app slug.

Figure 2 — Pipedream addresses its tool surface by app, Fabriq addresses it by user. Which one is better depends on whether your agent knows what it needs before it reads the request.

Neither is obviously right. Addressing by app keeps the tool list small, which is a real problem at 10,000 tools, and lets you hand a narrow server to a narrow sub-agent. Addressing by user means one connection and no router, at the cost of a tool list that grows with everything that user connected. If we had 10,000 tools we'd probably have built it Pipedream's way too.

OAuth and the Token

Both platforms default to their own OAuth client, which is the thing that makes a first call possible on a Friday rather than three weeks after you file for Google verification.

Pipedream's docs say it "will default to using Pipedream's official OAuth client", and that "for any OAuth app that supports it, you can always use your own client", with one hedge worth reading twice: "Your ability to use Pipedream's OAuth clients in production depends on the use case." The capability table is where that gets concrete, and it maps directly onto the question of who holds the token:

So on Pipedream's own client the raw credential never comes back to you. You run tools and proxy requests, and the token stays on their side. The moment you need the access token in your own process, you have to register your own OAuth app with the provider, create the client in Pipedream, and register the unique redirect_uri it returns. The client secret is write-only and never returned.

That is a well-drawn boundary and close to how we'd draw it. Fabriq's default proxy mode keeps the credential in the vault and attaches it server-side at call time, so the agent receives a session and never the secret, and our catalogue offers the same two paths per provider: Agentic Fabriq's OAuth app, or your own. Credential model and delegation model are two of the nine dimensions in the agent access control taxonomy, and on both of those the two products land in nearly the same place.

Where we differ is identity. Pipedream's MCP requests carry x-pd-external-user-id, a value your code sets per request. Fabriq's SDK takes your framework's request object and reads the user from the session your own auth already established, so there's no user id for your code, or a model inside your code, to get wrong. Each minted token also records an auth_level: app_asserted when we simply took your app's word, app_signed when your backend signed a 60-second single-use assertion with a private key we never hold, and user_present for the moments right after a human finished a connection. You can require signed assertions and refuse unsigned token requests. One thing we don't do: disconnecting clears the grant in our own vault and nothing else, so if you need Google to forget, ask Google.

Where Pipedream Wins

A comparison that concedes nothing isn't worth reading, so here is the honest list, and it's not short.

Reach, and it isn't close. Pipedream advertises 3,000+ APIs and 10,000+ prebuilt tools. Fabriq's catalogue holds 509 providers, and the composition matters: 123 have a live OAuth redirect, 12 are in progress, and 374 have no OAuth flow at all and are connected once in the console by pasting a credential.

End-user connect across the whole catalogue. This is the gap that would decide it for a lot of products. Pipedream's frontend SDK and Connect Link let your customer authorize anything in the catalogue. Fabriq's end-user connect flow currently drives thirteen providers end to end: Gmail, Google Drive, Docs, Sheets, Slides, Calendar, Meet, Forms, Contacts and Chat, plus Slack, GitHub and Notion. Our hosted credential-entry page, where a user pastes an API key, supports Notion only. Everything else is connected by you in the console, not by your user in a browser.

Free discovery. Listing and configuring costs no credits, so an agent can explore the catalogue without metering.

A prototyping environment with every feature on. All Connect features work in development on any plan, which is a better place to learn the product than a production tier with a ceiling.

Enterprise plumbing already built. Dedicated static IPs, SSO, SCIM and HIPAA on the Business plan, plus isError tool results carrying an attribution field that classifies whether a failure came from the tool's code, the upstream API or the network. That last detail is small and it's the kind of thing you miss at 3am.

What we'd claim in return is narrow. The free tier runs in production. The credential stays out of the agent's process by default. Identity comes from the request rather than a parameter. And the bill doesn't grow with connected users.

Frequently Asked Questions

Is Pipedream free? There's a free plan with all build features, 3,000+ connectors and 10,000+ prebuilt tools, but the pricing page marks the production environment and usage credits "Development only". Connect's development environment is capped at 10 external user IDs and intended for your own test accounts. Shipping to real customers means Startup at $99 a month.

Is Pipedream shutting down? No. Pipedream Workflows and String shut down on 31 March 2027, with Workflows data including connected-account credentials deleted by 30 April 2027. Pipedream Connect, the API, MCP servers and the proxy are explicitly unaffected and are where the company says its focus is going.

How do Pipedream credits work for an AI agent? A tool call through MCP, an action execution, a Connect proxy request or a deployed trigger emit each cost 1 credit per 30 seconds of compute. Startup includes 10,000 credits a month and additional credits are $0.012 each. Listing and configuring are free.

How many connected accounts does the free plan allow? Pipedream's docs say the free plan has a daily credit limit and a "limited number of connected accounts" but don't publish either figure. We're not going to estimate it. Check your billing settings, which the docs point to for your exact quota.

Do I need my own OAuth app? Not to start, on either platform. Pipedream defaults to its own OAuth client, and Fabriq defaults to ours. You need your own client on Pipedream if you want to retrieve user credentials through the API or invoke workflows as the end user.

Which has more integrations? Pipedream, by roughly six to one on providers. 3,000+ APIs against Fabriq's 509, and Pipedream's end-user connect flow covers its whole catalogue where ours covers thirteen providers.

What does Fabriq do that Pipedream doesn't? A production-capable free tier, no per-user billing line, credentials attached server-side so the agent never holds the secret, user identity read from your framework's request instead of a header your code sets, and an effective tool list computed as the intersection of the user's scopes and the agent's.

Conclusion

The interesting thing about this comparison is that it stopped being a comparison between a workflow product and an agent product. Pipedream decided that itself, in writing, with a date on it. What's left is two products trying to be the same thing from opposite directions, and they're priced as if they still had their old shapes.

Pipedream prices compute, because it used to sell compute. Fabriq prices requests, because it never sold anything else. If your agent makes a small number of long calls, Pipedream's meter is kinder. If it makes many short ones, which is what tool-using agents actually do, ours is. And if your product needs a thousand customers to each connect an obscure SaaS account this quarter, the meter is the wrong thing to be arguing about, because only one of these two catalogues can do it.

Sources

• Pipedream pricing — Free marked "Development only", Startup at $99 monthly or $90 annual with 100 users and 10,000 credits, "then $2 per user", additional credits "$0.012 each", Business custom and annual-only (checked October 2026)
• Pipedream plans and pricing docs — what consumes credits and what doesn't, the free plan's unpublished daily credit limit and connected-account cap, and billing on unique external users (checked October 2026)
• Workflows and String are shutting down on March 31, 2027 — the shutdown date, maintenance mode, the 30 April 2027 data deletion including connected-account credentials, and the Connect carve-out (checked October 2026)
• Pipedream Connect environments — the 10 external user ID cap in development, the same-browser sign-in requirement, and the production warning (checked October 2026)
• Develop with Pipedream MCP — the remote.mcp.pipedream.net/v3 base URL, the required x-pd-app-slug header, per-app servers, Connect Link returned in tool responses, and isError with attribution (checked October 2026)
• Pipedream OAuth clients — the default Pipedream client, the capability table showing credential retrieval requires your own client, and the write-only client secret (checked October 2026)
• Pipedream app discovery — apps.list with a query string for picking an app slug programmatically (checked October 2026)
• Pipedream Connect quickstart — the CLI path from install to a running SDK playground (checked October 2026)
• Agentic Fabriq pricing — Free forever at 20,000 calls a month with unlimited agents, Launch at $50 for 200,000 calls, Growth at $300 for 2,000,000, triggers from Launch upward, and what counts as a call
• Fabriq Developer and the Developer Quickstart — AF_APP_ID / AF_APP_SECRET, initiate_connection, for_user, 900-second user-scoped tokens, the auth_level ladder, the thirteen end-user OAuth providers and the Notion-only credential page
• Agent access control taxonomy — the credential and delegation model dimensions referenced above