The license is the point

Plenty of automation tools call themselves open source. Read the fine print and most are "source-available": you can see the code, but a license clause stops you doing much commercial with it. Activepieces is the real thing. Its core is MIT licensed, the permissive license with essentially no strings, which means you can self-host it, fork it, build your own integrations, and even resell what you build without asking anyone.

That matters if you have been burned by n8n's fair-code Sustainable Use License or by a "free" tool that turned out to gate the useful parts. With Activepieces the community edition is not a crippled demo of the paid product. It is the product, MIT and all, and that is the single most important fact about it.

What it does

Functionally it is a Zapier-style visual automation builder. You chain "pieces" (their word for integrations) into flows with triggers, actions, branches, and loops. The library sits around 680 to 700 pieces as of mid-2026, covering the mainstream apps most teams actually use, and because pieces are written in TypeScript in an open framework, adding a missing one is a pull request rather than a support ticket.

The builder is clean and quick to a first result, closer to Zapier's friendliness than to n8n's steeper canvas. For a team that wants automation it can host itself, without either the resale restriction or the developer-heavy learning curve, that middle ground is the pitch.

Pricing, and the model that changed

Activepieces changed how it charges, and if you read an older review you will see the wrong numbers. It used to meter by tasks (a free 1,000 tasks/month, then paid tiers around $25 and $150). As of a pricing update dated May 14, 2026, the cloud Standard plan works differently:

• Standard: free to start with up to 10 active flows, then $5 per active flow per month. Runs are unlimited, and AI agents, MCP servers, and tables are unlimited too.
• Ultimate: custom annual pricing, adding governance features: team and personal projects, piece-access controls, custom RBAC, SSO, and audit logs.

The shift from per-task to per-active-flow rewards heavy usage and punishes sprawl. If you have five flows that each fire thousands of times a day, unlimited runs at $25/month is a bargain. If you have sixty tidy little flows that barely run, you are now paying $300/month for automations that a task meter would have billed as pennies. Count your active flows before you assume this is cheap.

AI features run on credits, with cloud overage at $1 per 1,000 credits, and you can set a hard spending cap so an agent loop cannot run up a surprise bill. Self-hosting sidesteps the whole flow meter: the community edition has no flow limit and no task limit, so the only cost is your server, typically $5 to $20 a month.

AI agents and MCP

Activepieces has leaned into agents rather than treating AI as a single "call ChatGPT" action. You can build native AI agents that pick their own tools, and it acts as an MCP (Model Context Protocol) server, meaning you can expose your flows and connected apps as tools that an external assistant like Claude can call directly. On the paid cloud plans both AI agents and MCP servers are uncapped.

This is the area moving fastest, so treat specifics as a snapshot. The direction, though, is clear: Activepieces wants to be the open-source place you wire an LLM to your real apps, not just a workflow tool with a chatbot glued on.

Limitations

The integration library, around 700 pieces, is real but smaller than Zapier's 7,000-plus or Make's thousands. For common stacks you will be fine; for a niche B2B app, check the directory before you commit, because "write it yourself in TypeScript" is only comforting if you have a developer to hand.

The platform is also younger than the incumbents, so the ecosystem of tutorials, templates, and Stack Overflow answers is thinner. And the per-active-flow cloud pricing genuinely does not suit a many-small-flows setup. That is the case where self-hosting stops being a nice-to-have and becomes the sensible choice.

Verdict

Activepieces is the one to pick when the open-source license is not a checkbox but a requirement: you want to self-host without a resale clause hanging over you, keep your automation data on your own infrastructure, or build and ship custom pieces. It is friendlier than n8n for non-developers while giving up none of that freedom, and the MIT core is the cleanest license in this category.

Look elsewhere if you depend on a long tail of obscure integrations, where Zapier's sheer connector count still wins, or if you run dozens of low-traffic flows on cloud, where per-active-flow billing gets expensive fast (self-host, or use a task-metered tool). If you specifically want heavy in-workflow code and serverless-style compute, Pipedream or n8n fit that better.