Zapier SDK MCP Server

A MCP (Model Context Protocol) server implementation for the Zapier SDK, allowing AI assistants to interact with Zapier's APIs through a standardized protocol.

What is MCP?

The Model Context Protocol (MCP) is a protocol developed by Anthropic that enables AI assistants like Claude and Cursor to interact with external tools and data sources through a standardized interface. This package creates an MCP server that exposes all Zapier SDK functions as MCP tools.

Installation

This package is part of the Zapier SDK monorepo and is automatically installed when you install the CLI:

npm install -D @zapier/zapier-sdk-cli

Usage

Starting the MCP Server

You can start the MCP server using the CLI:

# Start the MCP server
npx zapier-sdk mcp

# Enable debug logging
npx zapier-sdk mcp --debug

Using with Claude Code

To use this MCP server with Claude Code, add it to your MCP configuration:

{
  "mcpServers": {
    "zapier-sdk": {
      "command": "npx",
      "args": ["zapier-sdk", "mcp"]
    }
  }
}

Programmatic Usage

You can also use the server programmatically:

import { createZapierMcpServer, startMcpServer } from "@zapier/zapier-sdk-mcp";

// Create and configure the server
const server = createZapierMcpServer({ debug: true });

// Or start it directly
await startMcpServer({ debug: true });

Stability

Every tool belongs to one of three stability tiers. Stable tools are exposed by default; opt into wider tiers with ZapierMcpServerOptions.stability, --stability <level> on zapier-sdk mcp, or ZAPIER_STABILITY=<level> in the environment.

| Tier | Contract | | ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Stable | The surface we intend to keep. While the SDK is in Open Beta (0.x), a breaking change can still land in a minor release; we deprecate first where we can, and every one is called out in the changelog with a migration note. | | Beta | Functionally complete and ready to build against, but the shape may still change as it settles. Changes land in minor releases with a changelog migration note, without a deprecation period. | | Experimental | Still being designed; may change shape or disappear in any release, without notice. |

Non-stable tool descriptions are prefixed with [Beta] / [Experimental] so agents see the tier inline.

How It Works

The MCP server:

1. Dynamically discovers all available Zapier SDK functions from the SDK's internal registry
2. Converts function schemas from Zod format (used internally by the SDK) to JSON Schema format (required by MCP)
3. Exposes functions as MCP tools with kebab-case naming (e.g., listApps becomes list-apps)
4. Handles tool execution by validating inputs, calling the appropriate SDK function, and returning results

Available Tools

All Zapier SDK functions are automatically exposed as MCP tools:

• list-apps - List available Zapier apps
• list-actions - List actions for a specific app
• get-action - Get details about a specific action
• run-action - Execute a Zapier action
• list-connections - List your connections
• And many more...

Error Handling

The server provides comprehensive error handling:

• Validation errors include the expected input schema
• Unknown tools list all available alternatives
• Execution errors provide context about which function failed
• Debug mode provides detailed error logging

Architecture

AI Assistant (Claude)
      ↓ MCP Protocol
MCP Server (@zapier/zapier-sdk-mcp)
      ↓ Function Calls
Zapier SDK (@zapier/zapier-sdk)
      ↓ HTTP Requests
Zapier APIs

The MCP server acts as an adapter, translating between the MCP protocol and the Zapier SDK's function-based interface.

Development

Building

pnpm build

Testing

pnpm test

Table of Contents

• Installation
• Usage
• Stability
• Available Tools
• Accounts
  • get-profile
• Actions
  • create-action-run
  • get-action
  • get-action-input-fields-schema
  • get-action-run
  • list-action-input-field-choices
  • list-action-input-fields
  • list-actions
  • run-action
• Apps
  • get-app
  • list-apps
• Client Credentials
  • create-client-credentials
  • delete-client-credentials
  • list-client-credentials
• Code Workflows (Experimental)
  • cancel-durable-run
  • create-workflow
  • create-workflow-draft
  • delete-workflow
  • disable-agentic-management
  • disable-workflow
  • discard-workflow-draft
  • enable-agentic-management
  • enable-workflow
  • get-agentic-management-config
  • get-agentic-management-intent
  • get-durable-run
  • get-trigger-run
  • get-workflow
  • get-workflow-draft
  • get-workflow-run
  • get-workflow-version
  • import-workflow
  • list-durable-runs
  • list-workflow-drafts
  • list-workflow-runs
  • list-workflow-versions
  • list-workflows
  • publish-workflow-draft
  • publish-workflow-version
  • run-durable
  • run-workflow-draft
  • trigger-workflow
  • update-agentic-management-config
  • update-agentic-management-intent
  • update-workflow
  • update-workflow-draft
  • validate-workflow
• Connections
  • create-connection
  • find-first-connection
  • find-unique-connection
  • get-connection
  • get-connection-start-url
  • list-connections
  • wait-for-new-connection
• HTTP Requests
  • fetch
• Human Inputs (Experimental)
  • create-form
• Tables
  • create-table
  • create-table-fields
  • create-table-records
  • delete-table
  • delete-table-fields
  • delete-table-records
  • get-table
  • get-table-record
  • list-table-fields
  • list-table-records
  • list-tables
  • update-table-records
• Triggers
  • ack-trigger-inbox-messages
  • create-trigger-inbox
  • delete-trigger-inbox
  • ensure-trigger-inbox
  • get-trigger-inbox
  • get-trigger-input-fields-schema
  • lease-trigger-inbox-messages
  • list-trigger-inbox-messages
  • list-trigger-inboxes
  • list-trigger-input-field-choices
  • list-trigger-input-fields
  • list-triggers
  • pause-trigger-inbox
  • release-trigger-inbox-messages
  • resume-trigger-inbox
  • update-trigger-inbox
• Virtual File System (Experimental)
  • append-vfs-file
  • copy-vfs-file
  • create-vfs-directory
  • delete-vfs-path
  • list-vfs-directory
  • move-vfs-path
  • read-vfs-file
  • stat-vfs-path
  • write-vfs-file

Available Tools

All SDK functions are automatically exposed as MCP tools using kebab-case naming.

Accounts

get-profile

Get current user's profile information

Actions

create-action-run

Start an action run and return its ID without waiting for the result. Running an action is asynchronous: this hands back a run ID immediately, and getActionRun fetches the outcome. Reach for this pair when you want to start work and collect it later (fan out many runs, hand the ID to another process, survive a restart). runAction is the one-call form that starts a run and waits for its result.

get-action

Get detailed information about a specific action

get-action-input-fields-schema

Get the JSON Schema representation of input fields for an action. Returns a JSON Schema object describing the structure, types, and validation rules for the action's input parameters.

get-action-run

Fetch the current state of an action run started by createActionRun. This is a point-in-time read that returns immediately: a run Zapier has not finished executing comes back with status waiting, so call again to check for a result. runAction starts a run and waits for its result in one call. Results are stored for seven days after the run was created.

list-action-input-field-choices

Get the available choices for a dynamic dropdown input field

list-action-input-fields

Get the input fields required for a specific action

list-actions

List all actions for a specific app

run-action

Execute an action with the given inputs and wait for its results. Each result page includes the action run ID in meta.id.

Apps

get-app

Get detailed information about a specific app

list-apps

List all available apps with optional filtering

Client Credentials

create-client-credentials

Create new client credentials for the authenticated user

delete-client-credentials

Delete client credentials by client ID

list-client-credentials

List client credentials for the authenticated user

Code Workflows (Experimental)

ℹ️ Experimental. Still being designed; may change shape or disappear in any release, without notice. Pass --stability experimental to zapier-sdk mcp, or set ZAPIER_STABILITY=experimental in the environment, to use these tools.

cancel-durable-run 🧪 experimental

Cancel a run-once durable run in initialized or started status. Returns 409 if the run is already terminal.

create-workflow 🧪 experimental

Create a durable workflow container. Starts disabled with no version; publish a version to add code.

create-workflow-draft 🧪 experimental

Fork a new draft from the workflow's current live version (or a blank stub before the first publish)

delete-workflow 🧪 experimental

Delete a durable workflow. Throws ZapierNotFoundError if the workflow doesn't exist; callers wanting idempotency should catch that themselves.

disable-agentic-management 🧪 experimental

Stop Agentic Management from working on one workflow. Its settings are kept, not deleted: the modes and auto-approve choice stay on file, so calling enableAgenticManagement later resumes with those settings instead of the defaults.

disable-workflow 🧪 experimental

Disable a durable workflow so it stops accepting triggers

discard-workflow-draft 🧪 experimental

Discard an open workflow draft (soft delete). The draft's unpublished edits stop resolving; the published version is untouched.

enable-agentic-management 🧪 experimental

Put one workflow under Agentic Management. A workflow disabled earlier comes back with the modes and auto-approve choice it had, unless you pass new ones here.

enable-workflow 🧪 experimental

Enable a durable workflow so it accepts triggers

get-agentic-management-config 🧪 experimental

Read the Agentic Management config for one workflow.

get-agentic-management-intent 🧪 experimental

Read the intent recorded for one workflow version: the plain-language description of what that version of the workflow is meant to do.

get-durable-run 🧪 experimental

Get the full state of a run-once durable run, including its operations journal

get-trigger-run 🧪 experimental

Get the workflow run associated with a deployed workflow's trigger. Useful immediately after firing a trigger, when you have the trigger ID but not yet the run ID.

get-workflow 🧪 experimental

Get a durable workflow with its current version details and trigger claim status

get-workflow-draft 🧪 experimental

Get full details of a workflow draft including source files

get-workflow-run 🧪 experimental

Get the current state of a workflow run — a live run of a published workflow version, or a draft run launched via runWorkflowDraft (the kind field says which)

get-workflow-version 🧪 experimental

Get full details of a workflow version including source files

import-workflow 🧪 experimental

Convert an existing Zap into a durable workflow. Transforms the Zap's latest published version (falling back to its draft only when the Zap has never been published) and puts the resulting source in an open draft. The workflow is created disabled with no published version, so publish the draft to make it live. The original Zap is untouched and keeps running.

list-durable-runs 🧪 experimental

List run-once durable runs for the authenticated account, newest first

list-workflow-drafts 🧪 experimental

List drafts for a workflow, most recently edited first (open drafts by default)

list-workflow-runs 🧪 experimental

List workflow runs for a specific workflow, newest first. Live runs (of published versions) and draft runs share the one list; pass kind to narrow it to one.

list-workflow-versions 🧪 experimental

List published versions for a workflow, newest first

list-workflows 🧪 experimental

List all active durable workflows for the authenticated account

publish-workflow-draft 🧪 experimental

Publish an open draft as a new immutable workflow version. Advances the workflow's live pointer and discards the draft — publish consumes it, so an open draft always means unpublished work. Continue editing by creating a new draft, which forks from the just-published version.

publish-workflow-version 🧪 experimental

Publish a new version of a durable workflow. Enables the workflow by default.

run-durable 🧪 experimental

Run a workflow's complete source-files map as a run-once durable run on code-substrate-runner (no deployed workflow required). Send every file the workflow needs, not only its entry point. Returns the run ID immediately; poll via getDurableRun for terminal status.

run-workflow-draft 🧪 experimental

Launch a draft test run: execute the draft's current source without publishing it. The draft's payload is snapshotted onto the run at creation, so later edits (or discarding the draft) never change what the run executed. Draft runs are free (not task-charged) and rate limited per user. The run is accepted asynchronously — it comes back initialized; follow it with getWorkflowRun, or list it via listWorkflowRuns with kind: "draft".

trigger-workflow 🧪 experimental

Look up a workflow's trigger URL and fire it manually, as the authenticated account.

update-agentic-management-config 🧪 experimental

Change the modes a workflow runs, whether fixes are applied without approval, or both. A modes list replaces the whole current set rather than adding to it.

update-agentic-management-intent 🧪 experimental

Record what one workflow version is meant to do. Writing again for the same version replaces the text already stored, so there is no separate create and update.

update-workflow 🧪 experimental

Update a durable workflow's name and/or description

update-workflow-draft 🧪 experimental

Update (autosave) an open workflow draft

validate-workflow 🧪 experimental

Validate durable workflow source without publishing it. Returns source diagnostics; issue kinds are extensible and may be unfamiliar to the client.

Connections

create-connection

Create a new app connection, end-to-end. Mints the start URL via get-connection-start-url, prints it to stderr, opportunistically opens it in a browser when it looks safe to do so (skipping CI / SSH / headless-Linux by default — pass --browser always to force, --browser never to suppress), then polls via wait-for-new-connection until the user completes OAuth and the new connection appears. Returns the connection.

This is the right command for most callers. Reach for the lower-level building blocks when you want either of: (a) hand off the URL and not block on completion — call get-connection-start-url alone, no wait-for-new-connection needed, or (b) do something custom between minting the URL and waiting — call get-connection-start-url, do your work (email or DM the URL, render a QR code, etc.), then wait-for-new-connection.

find-first-connection

Find the first connection matching the criteria

find-unique-connection

Find a unique connection matching the criteria

get-connection

Get details for a specific connection

get-connection-start-url

Mint a short-lived URL that begins an SDK-initiated connection flow. The URL is signed by zapier.com and bound to the current user/account — opening it in a different browser session will fail the binding check. Returns the URL as data so the caller decides what to do with it.

Use this directly (rather than the higher-level create-connection) when you want either of: (a) hand off the URL and not block waiting for completion — call this alone, skip wait-for-new-connection entirely, or (b) do something custom between minting the URL and waiting for the connection — call this, then email or DM the URL, render it as a QR code for mobile sign-in, etc., then call wait-for-new-connection. For the common case where you'd just print and poll back-to-back, create-connection is one call.

Pair with wait-for-new-connection to detect completion: pass the startedAt returned here straight through (it's the server's mint time, so polling isn't affected by client clock skew). Example (JS):

const {
  data: { url, app, startedAt },
} = await zapier.getConnectionStartUrl({ app: "slack" });
// hand `url` off — print it, DM it, email it, render a button, whatever
const { data: conn } = await zapier.waitForNewConnection({ app, startedAt });

list-connections

List available connections with optional filtering

wait-for-new-connection

Wait for a new connection to appear for the given app. Polls the connections list newest-first until the most recent matching row's date is at or after the started-at timestamp, then returns it. Pair with get-connection-start-url — that mints the URL the user opens, this waits for the resulting connection to land. Errors with a timeout after the configured timeout (default 5 min). Example (JS):

const {
  data: { url, app, startedAt },
} = await zapier.getConnectionStartUrl({ app: "slack" });
// show `url` to the user via the channel they're reading from
const { data: conn } = await zapier.waitForNewConnection({ app, startedAt });

HTTP Requests

fetch

Make authenticated HTTP requests to any API through Zapier. Pass a connectionId to automatically inject the user's stored credentials (OAuth tokens, API keys, etc.) into the outgoing request. Mirrors the native fetch(url, init?) signature with additional Zapier-specific options.

Human Inputs (Experimental)

ℹ️ Experimental. Still being designed; may change shape or disappear in any release, without notice. Pass --stability experimental to zapier-sdk mcp, or set ZAPIER_STABILITY=experimental in the environment, to use these tools.

create-form 🧪 experimental

Create a Zapier Form from a title and optional fields. Forms publish by default and return a shareable hosted_url; set draft to save without publishing.

Tables

create-table

Create a new table

create-table-fields

Create one or more fields in a table

create-table-records

Create one or more records in a table

delete-table

Delete a table by its ID

delete-table-fields

Delete one or more fields from a table

delete-table-records

Delete one or more records from a table

get-table

Get detailed information about a specific table

get-table-record

Get a single record from a table by ID

list-table-fields

List fields for a table

list-table-records

List records in a table with optional filtering and sorting

list-tables

List tables available to the authenticated user

update-table-records

Update one or more records in a table

Triggers

ack-trigger-inbox-messages

Acknowledge messages from a lease. Acked messages are removed from the inbox; unacked ones return to the available pool when the lease expires.

create-trigger-inbox

Create a new trigger inbox subscription. Always creates a new inbox; use ensureTriggerInbox for get-or-create on a stable key.

delete-trigger-inbox

Mark a trigger inbox for deletion

ensure-trigger-inbox

Get-or-create a trigger inbox by key. Idempotent on (user, account, key): returns the existing inbox if a matching subscription is registered, creates a new one otherwise. Throws ZapierConflictError if the key exists with a different subscription.

get-trigger-inbox

Get details of a trigger inbox by ID

get-trigger-input-fields-schema

Get the JSON Schema representation of input fields for a trigger. Returns a JSON Schema object describing the structure, types, and validation rules for the trigger's input parameters.

lease-trigger-inbox-messages

Lease up to N messages from a trigger inbox. Returns messages plus a lease ID; ack within the lease window to remove from the inbox.

list-trigger-inbox-messages

List messages in a trigger inbox (no payload, status-only)

list-trigger-inboxes

List all trigger inboxes for the authenticated user

list-trigger-input-field-choices

Get the available choices for a dynamic dropdown input field on a trigger

list-trigger-input-fields

Get the input fields required for a specific trigger

list-triggers

List all triggers for a specific app

pause-trigger-inbox

Pause a trigger inbox; events stop being collected

release-trigger-inbox-messages

Release messages from a lease back to the inbox without acknowledging them. Released messages become immediately available for re-leasing. The lease attempt still counts against the per-message lease limit; releasing does not refund the attempt.

resume-trigger-inbox

Resume a paused trigger inbox; events resume being collected

update-trigger-inbox

Update settings on an existing trigger inbox

Virtual File System (Experimental)

ℹ️ Experimental. Still being designed; may change shape or disappear in any release, without notice. Pass --stability experimental to zapier-sdk mcp, or set ZAPIER_STABILITY=experimental in the environment, to use these tools.

append-vfs-file 🧪 experimental

Append content to a file in the Zapier Virtual File System (VFS). Creates the file if it doesn't exist; missing parent directories are created automatically. Safer than reading the file and rewriting it — appends server-side without a read/write round-trip. Paths are absolute (e.g. /Mine/log.md).

copy-vfs-file 🧪 experimental

Copy a file to a new path in the Zapier Virtual File System (VFS). Both from and to are absolute file paths (e.g. /Mine/notes.md), not directories.

create-vfs-directory 🧪 experimental

Create a directory in the Zapier Virtual File System (VFS). Paths are absolute (e.g. /Mine/notes). Idempotent — safe to call on an existing directory.

delete-vfs-path 🧪 experimental

Delete a file or directory in the Zapier Virtual File System (VFS). Works on either; paths are absolute (e.g. /Mine/notes.md). Deletes are recoverable — contact Zapier support to restore an entry.

list-vfs-directory 🧪 experimental

List one page of a directory's contents in the Zapier Virtual File System (VFS). Paths are absolute (e.g. /Mine). Pass the previous page's nextCursor back in cursor to fetch the next page.

move-vfs-path 🧪 experimental

Move (rename) a file or directory in the Zapier Virtual File System (VFS). Both from and to are absolute paths (e.g. /Mine/notes.md). Fails if an entry already exists at to — delete or rename it first.

read-vfs-file 🧪 experimental

Read a file from the Zapier Virtual File System (VFS), the file store behind your context repository. Paths are absolute, e.g. /Mine/notes.md. Returns the content as a UTF-8 string with its size, version, and last-modified time.

stat-vfs-path 🧪 experimental

Inspect metadata for any path (file, directory, shortcut, or stream) in the Zapier Virtual File System (VFS). Paths are absolute (e.g. /Mine/notes.md). Check the type field before reading fields that only exist for certain types.

write-vfs-file 🧪 experimental

Write (create or overwrite) a file in the Zapier Virtual File System (VFS). Paths are absolute (e.g. /Mine/notes.md); missing parent directories are created automatically.