An open-source automation platform where flows, AI agents and 759 app integrations run on your own server, and every connected app doubles as an MCP tool.
What is
Activepieces
Activepieces is the automation platform you reach for when the license matters as much as the feature list. It is a visual builder where you chain steps called pieces into flows that move data between apps, run code, call APIs and hand work to an AI agent, and the core ships under MIT. The catalog returns 759 pieces today, each one a TypeScript npm package, and about 60 percent of them were contributed by users rather than the company. Depth varies more than that number suggests, since HubSpot carries 45 actions and 24 triggers while Apollo offers seven actions and no trigger at all. Above flows sit newer surfaces that arrived with the AI wave. Agents choose their own tools and can pause for human approval, tables hold records inside the platform, and the MCP server turns every app you have connected into a callable tool for Claude, Cursor or Windsurf. That MCP layer is the most distinctive thing on offer. We would keep building on n8n and reach for Activepieces in two cases, a legal team that refuses n8n's fair-code terms, or a client who wants automation embedded inside their own product. Nebor's work runs on n8n, and there is no Clay piece here, so the centre of our builds would need an HTTP step.
Integrations and the modern GTM stack
Activepieces calls its connectors pieces, and the public catalog API returned 759 of them on the day we checked. The counts below are the actions and triggers each piece actually exposes, since the headline number hides how uneven the depth is once you get past the popular apps.
HubSpot | The deepest CRM piece in the catalog with 45 actions and 24 triggers, covering contacts, companies, deals, tickets, lists and custom objects, plus triggers that fire on new or updated records. This is the one GTM connector we would trust to carry a full lifecycle without dropping to raw API calls. |
Salesforce, Pipedrive, Attio and Close | Salesforce runs 27 actions and 9 triggers, Attio 15 and 5, and the catalog carries 66 pieces tagged sales and CRM in total. Coverage is solid for create, update and search on standard objects, and anything unusual in your schema still goes through the HTTP step. |
Slack, Gmail and Google Workspace | Slack has 28 actions and 14 triggers, Google Sheets has 27 and 4, and Gmail is the third most used piece across the platform's own projects. These are the pieces most GTM alerting and hand-off flows are built from, and they are mature enough to be boring. |
Instantly, Smartlead and Lemlist | All three cold email platforms have official pieces, though they are thin. Instantly gives you 6 actions and 2 triggers, which covers pushing leads into a campaign and catching a reply, so anything past that means calling their API from the HTTP step. |
Clay | No Clay piece exists in the catalog. Clay itself sends outbound HTTP, so a Clay table can post into an Activepieces webhook trigger without trouble, and the return leg into a Clay table means an HTTP action with the token, headers and retry handling written by you. |
MCP clients | Connect your apps once in the Activepieces UI, then paste the MCP server URL into Claude Desktop, Cursor, Windsurf or a claude.ai connector, and the assistant can call those apps as tools. Tool categories can be switched on or off per project, which is the control you want before pointing a model at a live CRM. |
AI models | OpenAI, Azure OpenAI and Anthropic Claude all have official pieces, and 125 pieces sit in the AI category. Credits cover model calls at two to twenty per step, and supplying your own provider key drops that to one credit, which is the cheaper route once volume is real. |
Escape hatches | HTTP, Webhook, Schedule, Code and Postgres are core pieces, and the webhook and schedule triggers are the two most used objects on the platform. Half the catalog ships no trigger at all, so on 379 pieces these are the only way to start a flow. |
Pro’s
The MIT core self-hosts with no cap on runs, users or flows, so a client owns the automation outright and none of n8n's fair-code restrictions on resale or hosting apply.
Every app you connect is exposed as an MCP server, so Claude, Cursor or Windsurf can act on Gmail, HubSpot or Slack through credentials you authorised once inside Activepieces.
Credits meter the AI work rather than the automation, because a flow run costs one credit however many steps it contains, and bringing your own model key drops an AI step to a single credit.
Pieces are TypeScript npm packages with hot reloading, so a missing integration is an afternoon of building rather than a support ticket and a wait.
Flows keep a published version and a separate draft, so the live automation carries on running while you edit and you can fall back to the version that worked.
The vertical step builder reads more like a checklist than a canvas, and reviewers arriving from Zapier and n8n keep calling it the easier surface to hand to a non-technical colleague.
Cons
Half of the catalog cannot start a flow, because 379 of the 759 pieces ship actions and no trigger at all, so you fall back on a scheduled poll or a hand-wired webhook.
The MIT headline covers the flow engine, and agents, chat, projects, the API, SSO, audit logs and Git Sync sit under packages/ee on a commercial license, so a free self-host gets you little of the AI layer the marketing leads with.
There is no Clay piece in the catalog, so the enrichment layer at the centre of most GTM builds reaches Activepieces only through the HTTP step, with tokens, pagination and retries you maintain yourself.
The builder is a vertical list of steps, and reviewers describe a branch-heavy flow as a long scroll through nested containers where a canvas would let you spread the branches out and read them at once.
n8n's directory lists over 2,000 integrations against 759 pieces here, and its community has shipped far more templates and answers, so an unusual app or a strange error is more likely to be solved already over there.
Pricing
Community Edition
Free Self-hosted, MIT core, Docker or Kubernetes
No cap on runs, users or flows
Full piece catalog and the flow builder
Excludes agents, chat, projects and the API
No SSO, roles, audit logs or Git Sync
Community support only
Free
$0/month Cloud, one user, credits refresh daily
Unlimited flows
Flows, agents, tables and MCP servers
API access
Hard cap on credits with a daily refresh
Community support
Plus
$16/month Billed yearly, up to 5 users
Everything in Free
Three times the monthly credits
Pay as you go overage on credits
Bring your own AI keys
Community support
Team
$166/month Billed yearly, 25 users included, $25 per extra user
Everything in Plus
Five times the Plus credits
Unlimited projects
SSO, standard roles and global connections
Email support from the team
Ultimate
Custom Contact sales, yearly agreement
Everything in Team
Yearly credit pool and dedicated workers
Priority execution
SCIM, custom RBAC, audit logs and secret managers
Git Sync with releases and dedicated support
How it fits your stack
Alternatives
At a glance
759 pieces in the open catalog
MIT-licensed core, self-hosts free
AI agents with tool choice and approvals
MCP server for Claude, Cursor and Windsurf
Tables for storing records in-platform
Pieces are TypeScript npm packages
Free cloud tier with daily credit refresh
Embedding SDK for SaaS products
Read some of our client stories
GTM Workflows
Learn how to use GTM tools in workflows
Guides and tutorials to get the most from your workflows
Ready to get started?
By clicking Sign Up you're confirming that you agree with our Terms and Conditions.
