Skip to main content
The Agent SDK is for AI agents running in production that need to notify people as part of their own decisions. It gives your agent a set of ready-made SuprSend tools: trigger a workflow, read and update users, tenants and preferences, track events, and edit and publish workflows and templates.
This is distinct from the MCP server, which is for interactive AI-client sessions. The Agent SDK is for agents making decisions at runtime.
The Agent SDK is in early access and isn’t published as a package yet. It’s a Python toolkit for LangChain and LangGraph agents, and for agents that use OpenAI-style function calling. Join the waitlist to get access.

How it works

You define the notification system in SuprSend - workflows, templates, channels, preferences. The Agent SDK gives the agent tools that call SuprSend. To notify someone, the agent calls the trigger tool with a workflow slug and a payload, and SuprSend handles routing, delivery, and preferences.

When to use Agent SDK vs MCP


Guardrails

Permissions

Turn tool categories on or off for each agent, and block all writes with one switch. A second switch blocks destructive tools, which today covers only removing users from a list. An agent that only sends notifications can be limited to the trigger tool.

One workspace

Pin the agent to one workspace, so it can’t act on production by mistake.

Payload validation

Link a JSON schema to a workflow, and SuprSend rejects trigger payloads that don’t match it. See validate workflow payload.

Retries

Pass an idempotency key with a trigger and reuse it when you retry, and SuprSend won’t send the notification twice within 24 hours. Without one, the SDK sets a new random key on each call.

Logs

Agent-triggered notifications appear in your SuprSend logs like any other trigger, with the workflow, payload, routing, and delivery outcome.

Template governance

Notifications use your existing templates, so design and messaging stay consistent. Leave template editing out of the agent’s permissions to keep it that way. If you set no permissions, the agent gets every tool.

Example use cases

  • Document processing - Agent completes extraction and notifies the user with results via email.
  • Security monitoring - Agent detects anomalous login and triggers push and email alerts simultaneously.
  • Sales automation - Agent identifies buyer intent and routes a follow-up to the right channel.
  • Customer support - Agent resolves a ticket and sends the customer a summary via in-app and email.
  • Ops and incident response - Agent detects a delivery failure and escalates to on-call via Slack.

Get started

The Agent SDK is currently in early access.

Join the waitlist

Sign up for early access.

Book a demo

Talk to the team about your use case.

MCP Server

For development-time agent tooling.

Workflow trigger API

Trigger workflows from your own backend with the REST API.