Get started

almyty turns any API into tools your AI agents can call, served over MCP, A2A, UTCP, and Agent Skills. This page takes you from an empty account to your own API running inside Claude Code (or Cursor, or any MCP client) in a few minutes.

Do every step in the UI or from the CLI. Start on the hosted platform (no card required), or self-host with one container.


npm install -g @almyty/cli
almyty login

1. Turn an API into tools

Open APIs → Connect API, give it a name, and paste a schema URL. OpenAPI, GraphQL, SOAP, Protobuf, or an npm SDK all work. Leave Generate Tools checked.

Import an API

Every operation in the schema becomes a typed, validated tool. No code.

Tools generated from the schema

2. Publish a gateway

Tools reach your AI through a gateway. Open Gateways → Create Gateway, pick MCP, and assign the tools you just made from the Tool Scoping tab.

A gateway with tools assigned

One gateway is reachable over every protocol at once:

Protocol	For
MCP	Claude Code, Cursor, Windsurf, and other MCP clients
A2A	Google’s Agent-to-Agent orchestration
UTCP	Universal Tool Call Protocol
Agent Skills	30+ AI coding agents via SKILL.md

3. Call it from your AI (the payoff)

Point your assistant at the gateway and your API is live inside it. Grab the endpoint from the gateway’s Integrations tab (it looks like https://api.almyty.com/your-org/petstore) and a key from its Authentication panel. Then, in Claude Code:


claude mcp add petstore --transport http \
  --header "x-api-key: gw_live_..." \
  https://api.almyty.com/your-org/petstore

Your API's tools inside Claude Code

Ask Claude to do something and it calls your API directly:

Claude calling the gateway

If your client only speaks stdio, run the gateway as a local proxy instead:


npx @almyty/mcp-server your-org/petstore

That’s the whole loop: API → tools → gateway → your AI calling it.

Build an agent (optional)

Instead of handing tools to someone else’s AI, build your own agent on top of them. Compose LLM calls, tool calls, conditions, and loops on a visual canvas, or run an autonomous agent that figures out the steps itself, with cross-vendor verification, memory, and human approvals built in.

An agent built on your tools

Agents serve over the same protocols and an OpenAI-compatible endpoint, so anything that talks to OpenAI talks to your agent.

Where to next

Connect an LLM provider: bring your own keys, mix vendors
Drive almyty from the CLI: agents, chat, skills, runners, one login
Build agents: the visual builder and autonomous mode
Gateways in depth: auth, scoping, every protocol
Examples: end-to-end builds you can copy

Prefer to run almyty from your AI tools directly? almyty is itself an MCP server: see Control almyty via MCP.