Notiflows for AI

Notiflows is AI-native — manage notiflows as code and let your own AI agents inspect, create, and trigger them. Your AI, your tokens.

Notiflows is built so AI tools and agents can work with it natively. Every notiflow is an explicit, declarative definition an AI coding assistant can read, generate, review, and ship — and that your agents can trigger at runtime as a typed tool.

Your AI, your tokens

The guiding principle behind every AI surface: your AI, your tokens.

Each surface runs inside your AI tool or agent, on your model, paid for with your keys. Notiflows hosts no LLM and never runs a model on your behalf. The MCP server, agent toolkit, and skills are all thin clients over the Management API — they give your assistant the context and the tools to manage and trigger notiflows, but the intelligence (and the bill for it) stays with you. The only credential Notiflows needs is your account token, which authorizes API calls and is never exposed to the model.

This keeps the boundary clean: Notiflows is the notification platform; your AI is your AI.

Every surface authenticates with a single account token (prefix nf_at_) — create one in the dashboard and supply it to whichever surface you use.

The surfaces

MCP server

A hosted Model Context Protocol server that exposes the Management API as tools, so you can manage and trigger notiflows conversationally in Claude Code, Cursor, or Claude Desktop.

Agent toolkit

@notiflows/agent-toolkit — turns your published notiflows into typed tools for your own AI agents (Vercel AI SDK, OpenAI, LangChain).

Agent skills

Installable, progressive-disclosure skills that teach any compatible assistant the Notiflows schemas and CLI.

Notiflows as code (CLI)

@notiflows/cli — pull, edit, and push notiflows as local files your assistant can read and write directly.

llms.txt & docs for agents

Machine-readable docs — llms.txt, llms-full.txt, and raw-markdown doc routes for grounding an assistant.

How they fit together

The Management API is the source of truth for creating notiflows. The CLI and the MCP server are two interfaces over it — files-on-disk and conversational tools, respectively — sharing the same auth (an account token) and the same flat-steps notiflow shape. The agent toolkit is for the runtime side: giving your agents the ability to trigger notifications as function calls. And skills plus llms.txt ground any assistant in the real schemas so it produces valid notiflows on the first try.