Activepieces MCP Server

Activepieces is an open-source automation platform positioned as a self-hostable alternative to Zapier. Its MCP server turns the platform's 280+ integration "pieces" (Slack, Google Sheets, Stripe, Shopify, Notion, HubSpot, and many others) and its full flow-builder into tools that any MCP-compatible AI assistant can call. Instead of clicking through a visual builder, you can ask Claude or Cursor to compose, deploy, and inspect automation flows on your behalf.

The MCP server exposes capabilities across several categories: discovery (exploring flows, pieces, connections, tables, and runs), flow management (creating, duplicating, publishing, enabling/disabling), flow building (adding/updating/deleting steps and triggers), router and branching (conditional logic), tables (full CRUD on Activepieces tables), and testing/runs (executing flows and inspecting results). It is available both on Activepieces Cloud and on self-hosted instances.

Authentication is OAuth-based and is scoped to the authenticated Activepieces project, so user credentials and connection secrets for downstream apps stay inside Activepieces and are not exposed to the AI assistant. The server is reached over a remote SSE URL that you generate per project from the Activepieces dashboard.

Tools

Prerequisites

• An Activepieces account on cloud.activepieces.com or a self-hosted instance
• An MCP-compatible client: Claude Desktop, Cursor, Windsurf, or Claude.ai (custom connector)

Enable the MCP server

1. Open your Activepieces project.
2. Go to the project settings (gear icon) or to AI → MCP in the dashboard.
3. Toggle Enable MCP Server access on.
4. Copy the generated server URL. It looks like:

https://cloud.activepieces.com/api/v1/mcp/{SERVER_ID}/sse

Configure your client

Add the server URL to your MCP client config.

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "activepieces": {
      "url": "https://cloud.activepieces.com/api/v1/mcp/YOUR_SERVER_ID/sse"
    }
  }
}

Cursor (.cursor/mcp.json for project scope, ~/.cursor/mcp.json for global):

{
  "mcpServers": {
    "activepieces": {
      "url": "https://your-instance.com/api/v1/mcp/YOUR_SERVER_ID/sse"
    }
  }
}

Restart the client. On first connection, an OAuth flow opens in the browser to authorize the Activepieces project. After authorization, the client will list the available Activepieces tools.

Self-hosting

If you self-host Activepieces (see https://www.activepieces.com/docs/install/overview), replace cloud.activepieces.com with your instance hostname. The MCP endpoint follows the same /api/v1/mcp/{SERVER_ID}/sse path.

• Ask Claude to build a new flow end-to-end: "When a Stripe payment succeeds, append a row to Google Sheets and post a summary to Slack" and have it created, tested, and published in Activepieces.
• Triage failed runs by having the AI pull recent run history, identify error patterns across flows, and propose fixes to the offending steps.
• Manage data in Activepieces Tables from chat: query, insert, update, and delete rows without leaving the AI assistant.
• Refactor existing automations: ask the AI to add a router/branch to an existing flow so different customer tiers get different downstream actions.
• Onboard a new integration by asking the assistant to call ap_setup_guide for a piece (e.g. HubSpot) and walk through wiring up the connection.

• "Create a flow that watches a Gmail label for new emails, extracts the attachment, and uploads it to Google Drive."
• "Show me all flows that failed in the last 24 hours and summarize the most common error."
• "Add a branch to my lead-router flow so leads from companies with more than 200 employees go to the enterprise Slack channel."
• "List every piece I currently have a connection set up for, and tell me which ones are unused."
• "Insert a new row into the customers table with name 'Acme Inc' and plan 'Pro', then retry the latest failed run of the onboarding flow."

• Official, maintained by the Activepieces team and bundled with the open-source product.
• Broad surface area: 280+ app integrations plus full flow CRUD, tables, and run inspection.
• OAuth-scoped access keeps downstream app credentials inside Activepieces, not exposed to the LLM.
• Works with both Activepieces Cloud and self-hosted deployments.

• Requires standing up or signing up for an Activepieces project before the MCP server is useful.
• Public docs list tool categories but do not enumerate every individual tool name and schema, so discovery is mostly via the client's tool list at runtime.
• Remote SSE transport means clients that only support stdio MCP servers need a bridge like mcp-remote.

• n8n MCP server for an alternative open-source workflow automation platform.
• Zapier MCP for the hosted equivalent across Zapier's app catalog.
• Pipedream MCP for serverless workflow automation exposed via MCP.