The full stack
for AI agents.

Aerostack is the AI agent platform for developers — a full-stack environment for building AI-powered products on Cloudflare's global edge. From one dashboard you can host MCP servers (the tool layer that AI agents call), write and deploy edge functions with built-in database, cache, AI, queue, and storage bindings, compose agentic workflows on a visual canvas with 19 node types, publish intelligent bots for WhatsApp, Telegram, Discord, and Slack, and package reusable logic as skills that any agent can invoke. Everything — infra, auth, multi-channel delivery, human approval — is handled by the platform so you ship products, not plumbing. Claude, Cursor, ChatGPT, Gemini, Claude Code, CrewAI, and LangChain all work with Aerostack out of the box.

An MCP server exposes tools — database queries, API calls, business logic — to AI agents in a standard protocol they already speak. When you host an MCP server on Aerostack, you get edge deployment on Cloudflare Workers (near-zero cold start, globally distributed), a managed authentication layer, and the ability to compose many MCP servers behind a single workspace URL so agents don't juggle credentials per server. You can publish a server to the Aerostack marketplace and earn 90 percent of subscription revenue, or keep it private inside your workspace. Any agent that supports MCP — Claude, Cursor, Claude Code, and others — can discover and call your tools immediately.

A workspace is Aerostack's composition layer: it aggregates any number of MCP servers — yours, marketplace ones, or both — behind a single authenticated URL. An AI agent connects to that one URL and instantly has access to every tool across all the servers in the workspace, without you wiring separate integrations or managing credentials per connection. Workspaces also enforce human approval gates, so any tool call that crosses a policy line pauses and routes to a reviewer before executing. This is the surface Claude Code, Cursor, and other AI clients point at when they need real-world tool access. You can create multiple workspaces — one per project, team, or customer.

n8n and Zapier are deterministic pipeline builders: you hard-code a sequence of steps and they execute that sequence. Aerostack workflows are agentic — the LLM reasoning node is the decision-maker, choosing branches, calling tools dynamically from any MCP server in your workspace, and pausing for a human approval handoff when a step crosses a policy line. The 19 node types include identity verification, bot delegation, conditional splits, and subflows for reusable logic, all composable on a visual canvas with no YAML. If your automation has no reasoning step and no branching on model output — a fixed transform, a scheduled sync — a plain edge function is the right tool; reach for a workflow when you genuinely need an LLM in the loop.

Edge functions are single-purpose serverless handlers with built-in database, cache, AI, queue, and storage bindings — the unit of raw logic on the platform. MCP servers bundle one or more of those functions into a tool layer that AI agents can call. Skills are higher-level reusable packages — pre-built capabilities like a booking flow, a summarization chain, or a lead-qualification sequence — that any agent or workflow can invoke without rebuilding the underlying logic. Think of it as three layers: functions handle computation, MCP servers expose functions as agent-callable tools, and skills compose tools into repeatable agent behaviors. Aerostack lets you publish skills to the marketplace and earn 90 percent of revenue when other developers adopt them.

When you create a bot on Aerostack you configure it once — the agent logic, the tools it can call, the approval gates it respects — and deploy it across WhatsApp, Telegram, Discord, and Slack simultaneously. Each channel gets the same reasoning engine and workspace tool access; only the message formatting adapts per platform. Bots can delegate to each other: a WhatsApp bot can hand a sub-task to a Slack bot mid-conversation, then resume. The same agentic workflow engine powers both the bot logic and stand-alone workflow automations, so a human-approval handoff or a complex LLM-branching sequence works identically whether it is triggered by a chat message or a webhook.

The free tier includes 500K AI tokens per month plus access to all platform surfaces — MCP server hosting, edge functions, agentic workflows, bots, and workspaces. If you bring your own model key (BYOK) for Claude, OpenAI, or Gemini, the platform markup on AI calls drops to zero and the token meter reflects only what you consume at cost. Paid plans extend resource limits. For creators who publish MCP servers or skills to the Aerostack marketplace, the revenue share is 90 percent — the platform keeps 10 percent. There is no separate charge per channel for bots; WhatsApp, Telegram, Discord, and Slack all run under the same plan.

Aerostack is a platform for building AI products — it is not a managed-OAuth glue layer like Composio, not a generic LLM gateway, and not a no-code automation builder for non-technical users. If you need a click-and-point tool to connect SaaS apps with no code, n8n or Zapier is likely a better match. If you only need to proxy model API calls without building anything on top, a standalone LLM gateway is simpler. Aerostack is the right choice when you are a developer who wants to ship an AI-powered product — something with real backend logic, multi-channel delivery, agent-callable tools, and agentic workflows — without assembling that stack from scratch. The more your product looks like an agent platform itself, the more value you get from Aerostack.