Ship a real AI assistant on your SaaS in 30 minutes, not 3 weeks.

Embed a chat widget. Plug in your own MCP server (or fork one of ours). Your users talk to an AI that actually does things, because it can call the tools that matter to your product.

Get access → Read the 5-step guide → See it live →

How it works

Drop in the widget

One script tag on your page. Floating chat button, streaming responses, markdown, tool-call indicators.

Write an MCP server

A small HTTP server that exposes your product's actions as tools. Use @modelcontextprotocol/sdk, any stack. Our starter ships in a template repo.

Point Altex at it

Altex calls your MCP at runtime. Claude (or your model) sees your tools. Your users get an assistant that knows your product.

Quickstart

# 1. Embed the widget on your site
<script
  src="https://altex.ai/widget.js"
  data-api-url="https://api.altex.ai"
  data-api-key="ak_your_tenant_key"
  data-title="Ask your product"></script>

# 2. Spin up your MCP server (Next.js, Rails, FastAPI — any stack)
#    Starter template: github.com/omiga/altex-mcp-nextjs-starter

# 3. Email hello@altex.ai to get your API key.

Full walkthrough: altex.ai/docs →

Why not ___?

Why not vanilla Anthropic SDK?

You still have to wire streaming, per-user context, tool routing, rate limits, billing, a widget, and observability. Altex ships all of that. You write the tools.

Why not LangChain / CrewAI?

Every abstraction is a place where your agent can break at step 14 and you can't see why. Altex uses the MCP protocol, nothing proprietary. Your tools stay debuggable.

Why not Zapier AI?

Zapier's walled garden picks the tools for you. Altex lets you bring your own MCP server, so your AI can touch your actual product, not a generic integration.

Why not wait for the MCP ecosystem?

The protocol is ready. Anthropic, Cloudflare, Smithery, and OpenAI are all converging on it. Altex is the B2B platform layer you'd otherwise build yourself.

Running in production

Altex powers the AI assistant on omiga.com.au. Open the chat there to ask about services, products, or booking a consultation. The answer comes from the same Cloud Run + MCP stack you'd onboard onto.

omiga.com.au →

Who this is for

Solo or small-team technical founders building an AI assistant feature on a SaaS that already has users. Post-framework-rejection. You tried LangChain, ripped it out, and wrote your own with vanilla Anthropic SDK. You want the infrastructure without the abstraction.

Talk to us

We're onboarding the first 10 teams by hand. Fastest way in: email. We send an API key, walk through your first MCP call together, and confirm tools/list in your console before your users see anything.

hello@altex.ai →