Zapier MCP server: connect AI agents to Zapier from the EU

Zapier is an automation service from Zapier, Inc. in San Francisco that connects thousands of web apps through workflows called Zaps. Many teams already have their app accounts linked to Zapier.

Zapier MCP makes those connections available to AI agents. Its endpoint is https://mcp.zapier.com/api/v1/connect, reached over Streamable HTTP and authenticated with OAuth or a connection token. Rather than exposing every app at once, you add specific actions to an MCP server in Zapier, and only those actions appear as tools. This keeps the agent's options narrow and predictable.

Zapier is useful when the app you need has no MCP server of its own. A sales agent can add a lead to a niche CRM, a support agent can create a ticket in a help desk tool, and an office agent can add an entry to a shared calendar or a row to a spreadsheet.

Teams can also combine actions: an agent reads an enquiry, looks up the customer through a search action, then creates a task in a project tool and posts a note to the team channel.

In Zapier, create an MCP server, connect the app accounts the agent should use, and add the actions it may run. For each action, decide which fields the agent fills and which are fixed.

Then add the Zapier connector on inference.eu and connect with OAuth or paste the connection token. Credentials are stored encrypted in the EU and used only by the agents you allow. You can remove actions or revoke the connection in Zapier at any time; the app passwords and keys stay inside Zapier.

When an agent runs an action, Zapier, a US company, carries out the call to the target app, as it does for your Zaps. Data passes through Zapier and then goes to that app's API, so check where both handle it.

On inference.eu, the model that chooses actions and reads results runs in EU data centres, and your Zapier credentials are stored encrypted in the EU. Add only the actions an agent needs, and prefer fixed values for fields such as recipients, so personal data does not go where you did not intend.

Each Zapier action does something real in another app: it sends a message, creates a record or changes a file. Because one connector can reach many apps, a careless set-up can give an agent more reach than intended.

On inference.eu you can let search actions run on their own and require a person's approval before any action that writes, sends or deletes. The approval shows the action and the values the agent filled in. Fixed fields in Zapier add a second layer, so an agent cannot change a recipient or a target folder.