OneSignal MCP Server
OFFICIAL SERVEROneSignal's hosted MCP endpoint for push, email and SMS messaging, users, segments and exports
Hosted MCP server from Customer.io for working with its Journeys and Data Pipelines APIs
Official · customer.io Remote (hosted) · Streamable HTTP CRM and salesCommunicationData and analytics
Opens the app, which asks before it adds the server.
claude mcp add --transport http customer-io https://mcp.customer.io/mcphttps://mcp.customer.io/mcphttps://mcp.customer.io/mcp{
"mcpServers": {
"customer-io": {
"url": "https://mcp.customer.io/mcp"
}
}
}{
"servers": {
"customer-io": {
"type": "http",
"url": "https://mcp.customer.io/mcp"
}
}
}[mcp_servers.customer-io]
url = "https://mcp.customer.io/mcp"gemini mcp add --scope user --transport http customer-io https://mcp.customer.io/mcpCustomer.io MCP is the vendor's own hosted server. It connects an AI client to a Customer.io account, so the assistant can read workspace data and make changes: automations, one-time sends, segments, people, Design Studio content, CDP sources and delivery metrics. Nothing is installed locally. The client connects to https://mcp.customer.io/mcp (US) or https://mcp-eu.customer.io/mcp (EU) and signs in with OAuth using the user's own Customer.io login.
It does not have one tool per feature. It exposes eight generic tools that cover the whole Journeys UI API and the CDP Data Pipelines API. The assistant loads a reference (cio_prime), looks up endpoints and their parameters (cio_schema), and then calls the API through separate tools for GET, for POST/PUT/PATCH and for DELETE. Clients that allow per-tool approval can therefore let reads through and still ask before writes or deletes. Every API tool can do a dry-run preview first. Customer.io also serves a library of task instructions (skills), such as building an automation or setting up an SDK, which the assistant can list and read.
The setup page for IDEs says the endpoint uses streamable HTTP only and does not support the SSE transport. Request bodies are checked strictly: an unknown field returns a 422 error and is not silently dropped. Errors come back as structured JSON.
A Customer.io account where an account admin has turned on Customer.io MCP under Settings > AI. Customer.io AI does not need to be on. You also need an MCP client that can connect to a remote streamable-HTTP server and complete an OAuth sign-in in the browser. Gemini CLI needs `/mcp auth customerio` run after the server is added. EU accounts use the mcp-eu endpoint when the client is set up by hand.
A connection gets only the read scope by default; write, write:live and configure must each be requested and approved. Unless an admin turns on Allow MCP to edit live data, it cannot delete people or edit items in use, such as active automations or segments in use. Unless an admin turns on Allow MCP to access sensitive data (and the user has Sensitive Data View), it cannot read attributes marked sensitive. It can never do more than the signed-in user's role allows. SSE transport is not supported. Clients may cache the tool list and need a restart to see new tools. For terminal agents, Customer.io recommends its CLI instead.
Suits Customer.io users who want an assistant to look up campaign, segment and delivery data, or to draft automations and messages, from a chat client or IDE. There is nothing to run locally, and its OAuth scopes and admin switches let a team allow reads while gating sending and live edits. It reaches everything the API covers, so it fits people who are fine with the assistant choosing endpoints itself. If you work in Claude Code or another terminal agent, Customer.io's own docs point to its Customer.io CLI instead. It is of no use without a Customer.io account where an admin has enabled MCP.
cio_primecio_schemacio_read_apicio_write_apicio_delete_apicio_skills_listcio_skills_readcio_auth_status