AutomationConnector
Zapier
Run the Zapier AI Actions a workspace has exposed, preview them first and read each run back, from the same Claude or ChatGPT chat as the rest of your stack.
Official site: actions.zapier.com
What it can do
3 read
- 01Read
Run any action a workspace has exposed to Zapier's AI Actions
- 02Read
Preview an action before it runs
- 03Read
Read execution results
The connector reaches the actions a workspace has explicitly exposed in Zapier, not the whole Zapier automation surface. The agent lists those actions, runs one from a plain-language instruction, and reads the execution log afterwards. What it can trigger is fixed by whoever exposed the actions.
Does Zapier have an MCP server?
Yes. Zapier hosts its own MCP server and documents it at docs.zapier.com/mcp. It connects an MCP client such as Claude, ChatGPT or Cursor to the apps in your Zapier account, over Streamable HTTP (Server-Sent Events are not supported). Zapier keeps the app credentials and the OAuth handling on its side. A server runs in one of two modes: in agentic mode the model finds and adds actions during the conversation, and in managed mode you pick a fixed set. Tulina's connector is a narrower second route to Zapier, held in the same MCP link as your other tools.
How much does Zapier MCP cost?
Zapier says its MCP is included in every plan. Each successful tool call uses two tasks from your plan's task quota, the same quota your Zaps use, and failed calls do not count.
How do you connect Zapier to Claude and ChatGPT?
Through Tulina's MCP server, in three steps:
- Open actions.zapier.com, choose the actions to expose, and copy the API key. The set of exposed actions is attached to that key.
- Paste the key into Tulina's Zapier connector. It is stored in the vault and never shown to the model. A second Zapier account can be added under its own name.
- Add
https://mcp.tulina.ai/mcpto Claude or ChatGPT and sign in with your Tulina account. The MCP page has the same link.
Tulina vs Zapier's official MCP server
Both routes end in Zapier actions. They differ in who decides what the model can reach.
| Zapier's MCP server | Tulina | |
|---|---|---|
| Sign-in | OAuth from your MCP client, or a connection token for code | A Zapier API key in Tulina's vault |
| What the model reaches | Apps enabled on your server; in agentic mode the model can find and add more | Only the actions exposed to that key |
| Choosing an action | Discovery tools search for the right action | zapier_list_actions returns the exposed actions with their fields |
| Running an action | Tool call, two Zapier tasks per success | Action id plus an instruction, with optional field overrides |
| Dry run | Not covered on the pages we read | preview_only returns what would be done without running it |
| After the run | History of every MCP action in Zapier | zapier_execution_log returns one execution by its log id |
| Skills, code actions, Next Gen Zaps | Yes, per Zapier's docs | No |
Zapier's own server is the better choice when you want the model to discover actions across thousands of apps, save a workflow as a Skill or build a Zap from the chat. Tulina fits when the Zapier step belongs to a longer piece of work that also touches your inbox, CRM and calendar, and a person has already decided which actions are in play.
What you can ask it
"What can you do in Zapier?" The agent lists the exposed actions with their ids, descriptions and configurable fields, so you see the real menu before anything runs.
"Add this lead to the Zapier action that creates a Notion page, but show me first." The agent picks the action, fills the fields from your sentence and sets preview_only, so Zapier returns what it would do. You approve, and it runs the same call for real.
"Did that run succeed?" The agent reads the execution log for the id the run returned and reports the result, or the error, in plain words.
Zapier pairs with Slack when an action posts to a channel, and with HubSpot when the action is one a native connector does not cover. For the work those connectors do directly, see bug reports triaged and routed and customer call prep.
What the Zapier connector can't do
- It does not manage Zaps. It cannot create, edit, turn on or delete one, and it does not list your Zaps.
- It reaches only the actions exposed to the key. An action that is not exposed does not exist for the agent.
- Fields left in "AI guess" mode are filled by Zapier from the instruction. Pass
paramsfor any field that must be exact, and use the preview for anything that sends or writes. - The execution log is read by id. There is no call that lists past executions.
- No tool walks through several Zapier accounts at once. A total across accounts is one call per account.
Why through Tulina
The connection was never the hard part.
Most systems now publish their own way for a model to reach them, and that is good: it is what Tulina connects to. The hard part is deciding who may use it, what it is allowed to return, and what the model knows before it asks.
The keys stay in one place
YouTeamOrganisation
Connect Zapier once. The credential is encrypted in a vault, never held in clear, and never pasted into a chat or a workflow file. A person uses their own key where they have one, their team's where they do not, the organisation's after that, so nobody is blocked and nobody holds more access than they need. Revoke it once and everything running on it stops.
What leaves is filtered before it leaves
fields droppedmasked or hashedpseudonymised
Roles decide who can reach Zapier at all. Beyond that, a policy on the connector itself decides what comes back. The data is minimised at the source rather than after it has already reached a model.
Every call is on the record
whatwhowhen
Runs are logged for the organisation. That is what makes an automated action reviewable after the fact, and it is what a security review asks for first.
The run starts knowing your company
namingkey accountsdecisions
A direct connection gives a model access. It does not tell it how your company names things, which accounts matter, or what was decided last quarter. Tulina carries that across every tool, so the same request against Zapier returns something usable instead of something generic.
Processes that run on Zapier
None published yet.