# ManyMoney AI MCP server
The ManyMoney AI MCP server lets you drive your Pushwoosh project from the AI assistant you already use. Connect it to a [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) client such as Claude Desktop, Cursor, or Windsurf, and ManyMoney AI can plan, build, launch, and analyze Pushwoosh work on your behalf from natural-language prompts in your editor or chat.
If you have used [ManyMoney AI](/product/pushwoosh-ai/ai-assistant/) inside the Pushwoosh Control Panel, the MCP server gives you almost the same assistant outside the Control Panel.
## What you can do
Once you connect the MCP server, your AI client can work across the Pushwoosh feature areas you use every day.
| Area | What you can do |
| --- | --- |
| **Knowledge base** | Get answers about channels, journeys, integrations, and best practices grounded in the live Pushwoosh documentation instead of digging through the help center mid-task. |
| **Applications and account** | Launch a new app for a new market, add Android to an iOS-only project, issue an API token for a partner integration, or grant a teammate the right level of access without context-switching out of your chat. |
| **Audience and segmentation** | Build the segment you need on demand, like *users who bought in the last 30 days but haven't opened the app in 7*. Define the events you target on, tune frequency capping, and adjust the global control group as traffic ramps. |
| **Content** | Draft the push, email, SMS, and LINE presets for a campaign in one go, build rich-media in-app messages with personalized offers, and route email links through your own tracking domain so the brand stays consistent. |
| **Customer Journey** | Spin up a Welcome flow, an Abandoned Cart recovery, or a Win-Back series. Pause one that isn't pulling its weight. Clone a winner and adapt it for a new market, then pull the per-step funnel to see where users drop off. |
| **Diagnostics** | Stuck on *"why didn't this push land for these iOS users yesterday?"* Hand it to the assistant. It traces the message, correlates platform settings and the preset, and explains what to fix in plain English. |
| **Analytics and revenue** | Build the dashboard your CMO actually asks for, compare a journey's revenue lift against the global control group, and break a campaign down by app, country, or cohort without writing a single SQL query. |
| **Integrations** | Connect Stripe and start segmenting by payment events the same day, sync Meta for cross-channel retargeting, plug in Segment, Piano, Hubspot, Shopify, or a custom webhook, and check the connection status from the chat. |
| **Vouchers and rewards** | Drop ten thousand promo codes into a pool, wire it to your Welcome journey, and watch redemption stats land. This is handy for launches, win-back campaigns, and partner promotions. |
## Use cases
#### Build a campaign end-to-end from your editor
Ask your assistant: *"In application `XXXXX-XXXXX`, create a segment of users who added a product to the cart in the last 7 days but did not purchase. Then create a push preset 'Cart Reminder — discount 10%' and a Customer Journey that targets this segment, sends the preset 1 hour after entry, and exits if a `Purchase` event is recorded."*
Your assistant assembles the segment, the preset, and the journey. You review and publish the result in the Pushwoosh Control Panel.
#### Diagnose why a push did not arrive
While debugging in your AI client, ask: *"In application `XXXXX-XXXXX`, why did message `m-12345` fail for device `d-67890`? Explain in plain English."* The assistant investigates the delivery, correlates platform settings and the message preset, and explains the failure plus suggested fixes.
#### Analyze and report on revenue
Ask: *"Compare revenue and conversion for users who completed Journey `Welcome Flow` versus the global control group over the last 30 days, broken down by application."* The assistant assembles the comparison from your project's data so you can keep iterating on the analysis or paste the result into your weekly report.
#### Onboard a new project from a brief
Provide a short product brief and ask the assistant to *"create the application, add iOS and Android platforms, set up an Email From address, register a test device, create the standard Welcome and Cart Abandonment journeys with sensible defaults, and list everything you created so I can review."*
## Compatible AI clients
The ManyMoney AI MCP server works with any client that implements the MCP specification, including:
* **Claude Desktop** by Anthropic
* **Cursor** and **Windsurf**
* **Cline** and **Continue**
* **ChatGPT Desktop** with MCP support enabled
* Custom integrations built on top of the MCP specification
## Connect the MCP server
### Step 1. Make sure you have a Pushwoosh account
You need an active Pushwoosh account with access to the applications you want the assistant to work with. The MCP server runs every action under your Pushwoosh permissions, and your AI client can only do what you can do in the Control Panel.
### Step 2. Add the server to your AI client
Use the following endpoint in your client's MCP configuration:
```
https://manymoney.pushwoosh.com/mcp
```
Add the server to your Claude Desktop config file (`claude_desktop_config.json`):
```json
{
"mcpServers": {
"manymoney": {
"url": "https://manymoney.pushwoosh.com/mcp"
}
}
}
```
Restart Claude Desktop. On first use, the app opens a browser window so you can sign in to Pushwoosh.
Add the server to your `.cursor/mcp.json` (or the equivalent file in Windsurf):
```json
{
"mcpServers": {
"manymoney": {
"url": "https://manymoney.pushwoosh.com/mcp"
}
}
}
```
Reload the editor. The first request triggers a browser sign-in to Pushwoosh.
For any other MCP-compatible client, point it at `https://manymoney.pushwoosh.com/mcp` and follow the client's instructions for adding a remote MCP server.
### Step 3. Authenticate
The MCP server supports two authentication methods:
| Method | When to use it |
| --- | --- |
| **OAuth 2.0** | Recommended for desktop and editor clients. The client opens a browser, you sign in with your Pushwoosh account, and the token is stored locally. |
| **API token** | Recommended for headless or scripted setups. Send the header `Authorization: Token ` with every request. Create and manage API tokens in the Pushwoosh Control Panel. |
### Step 4. Try it out
Open a new chat in your AI client and ask something concrete and scoped to one of your applications:
> *In application `XXXXX-XXXXX`, list the customer journeys created in the last 30 days, and for each one show whether it is currently running and how many users entered it.*
If the connection works, your assistant pulls the data from your project and returns a grounded answer.
## Tips and best practices
* **Provide application codes and identifiers.** The same prompting tips that work in the [in-product ManyMoney AI](/product/pushwoosh-ai/ai-assistant/#prompting-strategies-and-best-practices-when-working-with-manymoney-ai) work over MCP. Reference applications by code, journeys by UUID, and segments by name to keep the assistant scoped.
* **Verify destructive actions.** Before approving a tool call that deletes data, archives a journey, or rewrites a segment, review the parameters. ManyMoney AI also asks for an explicit confirmation phrase for high-impact operations.
* **Keep the connection scoped per project.** Most MCP clients let you enable or disable individual servers per workspace. If you only want Pushwoosh available in a specific repo or document, configure it there.
* **Stay in your safety zone.** Sending live messages is intentionally not available over MCP. Plan, build, and review with the assistant. Trigger the actual broadcast from the Pushwoosh Control Panel or an automated journey.
## Data and your AI provider
The MCP server runs every action under your Pushwoosh permissions and applies the same rate limits and account scoping as the Control Panel. It does not, however, control what your AI client sends to its model provider. Your prompts, the tool arguments your assistant builds, and the Pushwoosh API responses it reads all flow through the AI client and the model vendor you have chosen — Anthropic, OpenAI, or whoever your client uses — and their privacy, retention, and training policies apply.
If you need Pushwoosh to handle the model side of the conversation under its own data handling rules, use [ManyMoney AI in the Control Panel](/product/pushwoosh-ai/ai-assistant/) instead. There, Pushwoosh routes prompts only through authorized subprocessors and applies the data handling described on the [ManyMoney AI page](/product/pushwoosh-ai/ai-assistant/#what-data-can-manymoney-ai-collect-and-process).
## Related