Arete documentation

Choose and Set Up an Agent

Arete works with any coding agent that can run commands and use MCP tools. Cursor, Claude Code, Codex, OpenCode, VS Code agents, and other supported clients all work. Nothing in the Arete workflow depends on one editor, so use the one you already like.

1. Open a project folder

Create or open the folder where you want the agent to work. We recommend setting Arete up per project. That way its dependencies, instructions, skills, and MCP configuration sit alongside everything else and can be reviewed with it.

mkdir my-arete-project
cd my-arete-project
git init

2. Give the agent the bootstrap page

Paste this prompt into the agent:

Agent prompt Copy / paste

Read https://docs.arete.run/agent.md and follow it to set up Arete in your environment. Verify the setup, and simply explain what Arete is capable of.

The agent will:

install the signed a4 CLI binary;
run a4 init -y;
run a4 doctor --json; and
follow the exact repair commands for any required check that fails.

You don’t need a Rust toolchain or an Arete account for basic setup and for searching the public catalog.

3. Reload the agent

a4 init writes MCP configuration for the agent it runs in and any agent set up in the project folder; add another with a4 init --agents <id>. Agents that run in a terminal often pick this up straight away. Desktop apps may need a full restart before they reload their configuration and the updated PATH.

Ask the agent to confirm that it can see:

the arete MCP server, for discovery and live views; and
the arete-docs MCP server, for documentation search.

If either one is missing:

a4 doctor --json
a4 doctor --fix

Review the changes it reports and restart the agent once more.

4. Ask an on-chain question

Swap your own question into the brackets:

Agent prompt Copy / paste

Use Arete to find programs and currently deployed views related to [YOUR QUESTION]. Inspect exact descriptors, gather only a bounded live sample, explain which programs and views support the answer, and disconnect when finished. If no suitable hosted view exists, tell me which program capabilities are closest instead of guessing.

Knowledge lookups and hosted connections may require you to sign in. When the agent tells you so, run:

a4 auth signup

or log in with a key you already control:

a4 auth login --key <a4_ak_...>

Don’t paste credentials into chat or into source code.

5. Turn the result into code

Once exploring has pinned down the right capability, ask the agent to install it and build on the generated types:

Agent prompt Copy / paste

Install the exact Arete program or stack we just validated into this project for [LANGUAGE]. Preserve arete.toml and arete.lock, inspect the generated SDK, and implement [FEATURE] using generated names and types. Do not edit generated files or reuse MCP cache data as an application dependency.

The agent uses a4 install stack ... when the application needs live views, with or without transactions: a stack includes the Program SDK for each program its views read from, the same one a standalone install gives you. It uses a4 install program ... when you only need program reads and transactions, and composes a stack when no single stack groups the views and programs you need.

What the setup writes

Depending on which agent it detects, a4 init may create or update:

arete.toml;
an Arete-managed block in AGENTS.md;
a CLAUDE.md bridge;
.agents/skills or an agent-specific skill directory;
project MCP configuration; and
skills-lock.json.

Running a4 init again updates the sections it manages and leaves the rest of your project instructions alone.

Safety expectations

An agent can search, read, generate local code, and inspect transactions without holding any authority over your transactions or deployments. Signing, submitting, publishing, and deploying are separate actions, and each needs an explicit request from you.

Continue with Explore On-chain with MCP or the Prompt Cookbook.