Create-fetch-agent

create-fetch-agent is an open-source scaffolding CLI that generates a runnable Fetch.ai uAgents project in one command — with unique seeds pre-filled, ports assigned, the Agent Chat Protocol already wired in, a Python environment set up, and AI-editor context installed.

Instead of cloning a starter, hand-wiring the chat protocol, generating a seed, and fighting dependency setup, you answer a few prompts and get an agent that runs immediately and is ASI:One-ready. The only things left as TODO are the workflow functions where your logic goes.

One command to get started

From anywhere on your machine:

npx create-fetch-agent

This launches the interactive wizard. No global install, no config files.

Why use create-fetch-agent?

• Runnable in one command — the generated project starts with make run; no manual seed, protocol, or environment wiring.
• Chat protocol always wired — every generated agent publishes its chat protocol manifest on startup, so it's discoverable and chattable on ASI:One. This is the #1 thing builders forget, and it's done for you.
• Batteries included — pre-generated unique seeds, deterministic ports, a Makefile, and a real dependency manifest for your package manager.
• Multiple project shapes — a single agent, a chat agent, an ASI:One-routed multi-agent project, or a pay-to-use agent that charges in on-chain FET.
• AI-editor context built in — installs Fetch-skills context for Cursor, Claude Code, Google Antigravity, or AGENTS.md, so your coding assistant extends the project correctly.
• You own deployment — nothing is auto-deployed. The CLI prints honest, step-by-step guidance for connecting your agents to Agentverse when you're ready.

Who is it for?

create-fetch-agent is built for developers building on the Fetch.ai ecosystem — and especially for hackathon builders who want to skip setup and get an agent onto ASI:One fast. If you're writing uAgents, splitting work across multiple agents, or building a pay-to-use service, it gives you a correct, runnable starting point in seconds.

Quick Start

npm install vs npx

Running npm install create-fetch-agent only adds the package to node_modules — it does not launch the scaffolder. Always use npx create-fetch-agent (or npm create fetch-agent) to run it.

Pass a project name (or omit it and the wizard will ask):

npx create-fetch-agent my-agent

The wizard asks a few questions with arrow keys and Enter, then generates the project, installs your chosen AI-editor context, and (optionally) bootstraps the Python environment.

Build types

The first prompt — What are you building? — picks the shape of your project:

Build type	What you get
Single agent	One self-contained, chat-enabled agent.
Chat agent (ASI:One ready)	A chat agent tuned to be discoverable and chattable on ASI:One out of the box.
Multiple agents (ASI:One routes between them)	Several independent expert agents — no orchestrator; ASI:One discovers and routes to each by its description.
Payment agent (FET)	A pay-to-use agent that charges in on-chain FET before running its paid action.

See Build Types & Flags for the file layout and run commands for each shape.

The wizard, step by step

Step 1 — Choose your Python setup

Pick the package manager for the generated project. The CLI writes a real manifest for whichever you choose — a PEP 621 pyproject.toml for uv, a Poetry pyproject.toml, or a requirements.txt for pip — and a Makefile that uses the matching interpreter.

Which one should I pick?

uv is the recommended default — it's fast and manages the virtual environment for you.

Step 2 — Add AI-editor context

Select any AI coding tools you use. The CLI installs Fetch-skills context into the generated project so your assistant knows the correct uAgents and protocol patterns. Use Space to toggle, Enter to confirm, or select nothing to skip.

Step 3 — Connect on Agentverse

The last prompts let you install dependencies now and choose whether to see Agentverse steps immediately. create-fetch-agent never auto-deploys — instead it prints exactly how to connect your agent, and confirms the chat protocol is already wired for you.

Run your agent

cd into the project and start it:

make run

The agent logs its address, publishes its chat protocol manifest, registers on the Almanac API, and prints an Agentverse inspector URL. Open that URL, click Connect → Mailbox, and your agent is live on ASI:One.

Testnet funds warning is normal

On startup you may see a not enough funds to register on Almanac contract warning. This is just the testnet faucet — it does not stop your agent from running or from being chattable. The Manifest published successfully: AgentChatProtocol and Registration on Almanac API successful lines are what matter.

Supported AI-editor targets

Tool	Install path
Cursor	.cursor/skills/<skill>/SKILL.md
Claude Code	.claude/skills/<skill>/SKILL.md
Google Antigravity	.agent/skills/<skill>/SKILL.md
AGENTS.md / generic	AGENTS.md

How create-fetch-agent uses fetch-skills

create-fetch-agent and Fetch-skills are complementary. create-fetch-agent generates a full, runnable project (code, seeds, ports, environment, protocols), and for the "AI-editor context" step it uses Fetch-skills under the hood to install the right SKILL.md files for your chosen tools.

	create-fetch-agent	Fetch-skills
What it does	Scaffolds a runnable uAgents project and installs AI-editor context	Installs SKILL.md AI-editor context only
Output	Runnable code, seeds, ports, .env, Makefile, protocols	SKILL.md files (no code)
Best for	Starting a new agent project from zero	Adding Fetch.ai know-how to an existing (or any) project

Already have a project?

If you only want to add AI-editor context to an existing project — exactly what Fetch-skills does — create-fetch-agent can do that too, without scaffolding:

npx create-fetch-agent --skills-only --ai cursor,claude

Frequently Asked Questions

Do I need to install anything globally? No. npx create-fetch-agent downloads and runs the latest version on demand.

Does it deploy my agent for me? No — by design. It generates a runnable project and prints the exact steps to connect on Agentverse. You stay in control of deployment.

Do I have to configure the chat protocol? No. Every generated agent already publishes its chat protocol manifest on startup, so it's ASI:One-ready out of the box.

Can I run it without answering prompts? Yes. Pass flags such as --type, --python, and --ai for a non-interactive run. See Build Types & Flags.

Does the payment agent need API keys? No. The payment agent charges in on-chain FET and defaults to the stable-testnet with a self-funding wallet — there are no payment keys to configure.

Is create-fetch-agent free to use? Yes. create-fetch-agent is MIT licensed.

Links

• npm package: npmjs.com/package/create-fetch-agent

Related docs

• Build Types & Flags — every project shape and CLI flag in detail
• Fetch-skills — the AI-editor context installer create-fetch-agent uses
• uAgent Creation — how to build and run a uAgent
• Agent Chat Protocol — the protocol wired into every generated agent
• Agent Payment Protocol — the payment flow used by the payment agent
• Agentverse — the platform your agents deploy and discover on