SuprSend Lets Claude and ChatGPT Run Its Notification Workspace

SuprSend said Oct. 12 that its MCP server is now hosted and in beta. Chat apps such as Claude.ai and ChatGPT can connect to a SuprSend workspace with one URL. The earlier server needed a CLI install and service-token authentication. That kept it out of chat apps. SuprSend announced the change in its product changelog.

Setup is one URL and a sign-in

Connecting takes the URL https://mcp.suprsend.com/mcp and a SuprSend sign-in, SuprSend said. No install, token or developer handoff is needed. SuprSend lists setup guides for Claude, ChatGPT and Codex, Cursor, VS Code with Copilot, Gemini CLI, Devin, Kiro, Replit, Lovable, Figma Make and Grok.

The server exposes 92 tools in 18 groups. Four skills come with the connection, covering SuprSend’s semantic layer, template schema, workflow schema and docs. The agent learns those schemas before it writes anything.

Agents can build, debug and send

A coding agent can add the SDK, identify users and build the in-app inbox or preference center from the workspace’s categories and channels. SuprSend’s example prompt is a cart-abandonment flow that waits an hour, then emails and sends an in-app message.

The agent can explain a missed notification. It checks delivery data and opens the message that went out, with links off and codes masked. Then it says where delivery stopped. It can update users, lists, tenants and preferences, send a template test, run a workflow for a test user or trigger a broadcast to a list.

Workflows and templates save as drafts, SuprSend said. They go live when the agent is told to publish.

Charts render inside the chat

The server is built on the MCP spec dated July 28, 2026, SuprSend said. Clients that support it show interactive results. The data_query tool returns a line, bar, stacked bar, pie or metric chart, or a table, with a Refresh button in the conversation. Message previews and interactive UI for workflows and templates are listed as coming soon. Clients built on the 2025 spec still connect and get plain text.

Guardrails limit what an agent can reach

Role-based access gives members read-only tools, SuprSend said. Only admins see tools that change data or send. Write and send tools are marked. The AI client asks for approval first. SuprSend advises against auto-approving sends.

The workflows_trigger tool carries an idempotency key. A retry within 24 hours sends nothing again. Adding a workspace to the URL, such as /mcp/staging, pins the agent to that workspace and keeps it out of production.

Teams already on the CLI-based server can swap its command block for the hosted URL. SuprSend said the CLI still fits CI/CD and syncing assets across workspaces.

The take

We think the guardrails matter more here than the 92 tools. An agent that can send notifications can reach customers. The question is what it can do before a person says yes. SuprSend’s answer is sensible: drafts by default, approval on sends and a retry that cannot double-send.

Those protections only work if people use them. An approval prompt clicked through on reflex protects nothing, and the pinned staging URL is optional. We’d start every agent on a staging workspace with a member role, and leave auto-approve off for write tools. Then trigger the same send twice and confirm the second does nothing.