ActivePieces is an open-source, TypeScript-based no-code automation platform for building AI workflows and exposing integrations as MCP servers, aimed at both technical and non-technical users who want Zapier-style automation while keeping control of their own data.

What it is

ActivePieces describes itself as an open source replacement for Zapier. It is an all-in-one AI automation platform built around an extensible, type-safe "pieces" framework written in TypeScript, where a piece is the unit of integration — one connector to one service. Pieces are npm packages, so the project lives squarely in the JavaScript and TypeScript ecosystem, and its integration source code ships as versioned packages published to npmjs.com. The platform pairs that developer-facing framework with a no-code builder, so people who do not write code can assemble the same pieces into working flows.

The concrete problem it solves is lock-in and data exposure in hosted automation SaaS. Instead of subscriptions and connectors that live only on a vendor's servers, ActivePieces lets a team run the whole automation stack itself, including full self-hosted and network-gapped deployments, and lets it extend coverage by writing or installing pieces rather than waiting on a vendor. Around 60% of the available pieces come from the community, and every contributed piece becomes automatically available as an MCP server, which turns the same integration library into tooling for LLMs.

Key capabilities

• Type-safe TypeScript pieces framework with hot reloading for local piece development, so contributors can build and test integrations on their own machine.
• Automatic MCP exposure: all pieces (280+) are usable as MCP servers with LLMs through Claude Desktop, Cursor or Windsurf.
• 200+ prebuilt pieces covering services such as Google Sheets, OpenAI, Discord and RSS, published to npm under the @activepieces scope.
• Builder features including Loops, Branches, Auto Retries, HTTP, and a Code with NPM step that supports "ASK AI in Code Piece".
• Human-in-the-loop controls that delay execution for a period of time or require approval, plus built-in human input triggers such as "Chat Interface" and "Form Interface".
• Fully versioned flows, language translations, and customizable templates.
• AI-first tooling: native AI pieces for experimenting with providers, and an AI SDK for building flows inside the builder.

Who uses it and how

• Enterprises where developers set up the tools and anyone in the organization then uses the no-code builder, with customization from branding through to control.
• Security-sensitive and network-gapped environments that need self-hosted automation and want to keep data inside their own perimeter.
• AI agent builders who connect pieces to Claude Desktop, Cursor or Windsurf as MCP servers instead of writing bespoke tool wrappers.
• Approval-heavy processes such as reviews and sign-offs, using delayed execution, approval steps, and form or chat triggers to bring a person into the flow.
• Open-source contributors, from integration authors publishing new pieces to non-coders contributing translations through the i18n workflow and the Discord community.

Getting started

Use the hosted option at activepieces.com, or self-host the Community Edition; pieces install as npm packages, and new integrations follow the Contributor's Guide for building pieces.

How it compares

Zapier is the paid product this project explicitly positions itself against, and n8n appears in the topic list as a comparable tool. The contrast rests on licence and hosting: the Community Edition is released under the MIT license while enterprise features sit under a Commercial License, so a team can self-host and network-gap the platform rather than depend on a vendor's hosted service while its data stays in that vendor's hands. The cost model follows from that split, with the pricing page documenting the feature comparison between the two editions.

When to use it — and when not to

A self-hoster takes on operating the instance and its data, which the project supports but does not eliminate; a team that wants zero-ops automation with no infrastructure should choose a hosted service instead. The licence situation needs checking up front, because GitHub reports it as NOASSERTION even though the README states MIT for the Community Edition and a Commercial License for enterprise features. The README excerpt also gives no install commands, image names or version numbers, so expect to lean on the docs site and the product changelog rather than the repository front page.