zapier/zapier-mcp: a plugin wrapper around a hosted MCP endpoint

What this repo is, and what it is not

The description on the repository is precise: it is the official plugin distribution for the hosted Zapier MCP server. The README states that the repo distributes the plugin that lives in the AI client, and the only content directory listed is plugins/zapier/. Everything else in the tree is metadata and contributor material: AGENTS.md, CONTRIBUTING.md, llms.txt, gemini-extension.json, assets, and a LICENSE file. There is no server source tree here. If you arrived expecting to run Zapier's integration layer yourself, you will not find it. The endpoint your client talks to is https://mcp.zapier.com/api/v1/connect, operated by Zapier, and the plugin is the thin piece that teaches your assistant how to use it well. That distinction matters for procurement and for anyone evaluating the project on the assumption that a MIT-licensed repository means a MIT-licensed server. It does not. The licence covers the plugin files in this repo, not the hosted service behind the URL.

The problem: MCP clients that do not know how to use Zapier

Connecting an MCP client to Zapier is a one-line configuration change. The README says connecting only takes a server URL, and points to docs.zapier.com/mcp/clients for per-client steps. What a bare connection does not give you is context. The assistant receives a tool surface covering, per the README, 9,000+ apps, and has to work out which action to call and how to phrase the arguments. The plugin distributed here exists to close that gap. According to the README it layers guided onboarding, a first-use demo, and skills tailored to your role on top of the base connection. So the audience is narrow and specific: teams already using an MCP-compatible client who find the raw connection too open-ended, or who want a repeatable first-run experience rather than each engineer improvising their way through the tool list. It is not aimed at people building their own agent framework, and it is not aimed at people who want to script Zapier without an AI client in the loop, since that is what Zapier's existing automation product already does.

How the connection and the plugin layer fit together

There are two layers, and the README keeps them separate. The base layer is an HTTP MCP server at https://mcp.zapier.com/api/v1/connect. The install badges in the README encode exactly that: the VS Code badge carries a config object with type set to http and url set to that endpoint, and the Cursor badge carries a base64 blob that decodes to the same URL. The second layer is the plugin from this repository, which the README says some clients install on top of the base connection. Installation of the plugin is per-client and goes through each client's own distribution channel rather than through a single command. Claude Code users add either zapier/marketplace or anthropics/claude-plugins-official. Codex and GitHub Copilot CLI users add zapier/marketplace. Claude Cowork users add anthropics/knowledge-work-plugins. Cursor users either use the badge or add cursor/mcp-servers. Kiro installs through kiro.dev/powers. Gemini CLI is the one client with a command in the README. VS Code is the exception in the other direction: the README states it connects directly to the hosted MCP server rather than through a plugin, which means the onboarding and skills described above are simply not part of the VS Code path.

Installing it: the one command in the README

For Gemini CLI the README gives a two-step sequence. First, gemini extensions install https://github.com/zapier/zapier-mcp. Then, /mcp auth zapier. The second step is the one people skip, and it is the step that actually binds your Zapier account to the client. For VS Code, the manual configuration is a server entry with type: http and the URL https://mcp.zapier.com/api/v1/connect, or the install badge if you prefer the redirect flow. For every other supported client, the README defers to docs.zapier.com/mcp/clients rather than reproducing the steps, and for plugin installation it names the marketplace repository you add, not a command. That is a real friction point: five of the listed clients are onboarded by adding a third-party marketplace repository to your client, which is a different trust decision than running an install command you can read. Before you do that, look at plugins/zapier/ in this repo to see what the plugin actually contains, and at gemini-extension.json to see how the Gemini CLI extension is declared. Those two paths are the only ones where the artifact is visible in this repository.

Where the plugin adds nothing, and where it can fail

The clearest limitation is stated by the README itself: VS Code connects directly to the hosted server rather than through a plugin. If your team standardises on VS Code, this repository is irrelevant to you and the badges are the whole story. The second limitation is structural. Because the plugin is distributed through client marketplaces and through third-party repositories such as zapier/marketplace, anthropics/claude-plugins-official, cursor/mcp-servers and anthropics/knowledge-work-plugins, an update to the plugin does not necessarily reach you when this repository changes. Last push on main was 2026-07-29, and no releases were retrieved, so there is no version number here to pin against or to compare with what your client has cached. The third is the one that bites in practice: every path routes through Zapier's hosted endpoint, so an outage or a credential problem at mcp.zapier.com looks identical from inside your editor to a broken plugin, and the plugin cannot help you diagnose it. There is no local fallback and no self-hosted mode described in the material. If your requirement is that tool calls execute inside your own network boundary, this project does not meet it and no configuration option in the README changes that.

The alternative, and the actual difference

The obvious alternative is to skip the plugin and connect your MCP client directly to https://mcp.zapier.com/api/v1/connect, which is what the VS Code instructions already do and what the Cursor and VS Code badges encode. The difference is not in reach. Both paths reach the same hosted server and the same catalogue of apps. The difference is what the assistant knows before the first call. Direct connection gives the model the raw tool surface and nothing else; the plugin adds guided onboarding, a first-use demo, and role-tailored skills, per the README. If your engineers are comfortable reading a tool list and picking actions, the direct connection is fewer moving parts, one less third-party repository in your supply chain, and no marketplace update to track. If your team is large, or if the first-run experience is the thing that determines whether people keep using it, the plugin is the part that does that work. A second alternative, for anyone whose real goal is scheduled or event-driven automation rather than conversational tool calls, is to use Zapier's automation product directly and leave MCP out of it. The README frames MCP as governed access from an AI client, and if no AI client is in the loop, the plugin has no role.

Maintenance cost and the licence boundary

The repository is MIT licensed, and the LICENSE file sits at the root. That covers the plugin code in plugins/zapier/ and the extension manifest gemini-extension.json. It does not cover the hosted service the plugin connects to, and nothing in the README suggests otherwise. Do not read the MIT badge as a statement about the server. On maintenance, the material supports only a limited answer. There are no releases retrieved, so there is no changelog to diff and no version to pin. The contribution path is documented in CONTRIBUTING.md, and AGENTS.md is described as a guide for AI agents working in the repo, which tells you the maintainers expect agent-assisted contributions rather than a large human contributor base. Practically, your upgrade cost is the cost of re-adding or refreshing a marketplace entry in each client your team uses, multiplied by the number of clients. If you run Claude Code and Gemini CLI side by side, that is two separate update paths for the same plugin. There is no single command in this repository that updates all of them.

Editorial conclusion

Adopt the plugin if you already use Claude Code, Cursor, Codex, Copilot CLI, Claude Cowork, Kiro or Gemini CLI and want guided onboarding plus role-tailored skills for Zapier MCP. Skip it if you are on VS Code, since that client connects straight to the hosted server, or if you need a self-hosted, auditable tool implementation, because this repo contains no server code. Before rolling it out, check plugins/zapier/ and gemini-extension.json to confirm what the plugin actually adds, and read the Zapier MCP client docs for the auth step your client requires.

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. zapier/zapier-mcp on GitHub