Skip to content

Strands harness quickstart

Strands harness is a state-of-the-art, fully assembled agent harness. You can get started in a few different ways:

Set up with your coding agent

Section titled “Set up with your coding agent”
Using a coding agent? Copy this prompt into Codex, Claude Code, Kiro, or any coding assistant and it will walk you through this page, ask which path, language, and model provider you want, and offer to set up the Strands MCP server.

Build an agent with the CLI

Section titled “Build an agent with the CLI”

The strands CLI’s setup assistant builds a custom agent by asking you a few questions — no code required.

Install the CLI

Section titled “Install the CLI”
Terminal window
npm install -g @strands-agents/cli

Create your agent

Section titled “Create your agent”

Run strands --setup (or /setup inside a chat) to open the setup assistant. It walks you through the creation process, asking:

  • Where to start — from scratch, or from a configuration it detects in the current directory.
  • Name and goals — what the agent is called and what you want it to do.
  • Capabilities — it recommends a model, built-in tools, Agent Skills, MCP servers, long-term memory, context management, and a tool-approval mode; accept its picks or adjust each one.
  • Appearance — theme and color mode.

It keeps the configuration as a draft until you approve it, then opens your custom agent in a fresh chat. For example:

$ strands --setup
? Start from scratch, or use the configuration in ./ ?   Start from scratch
? What should your agent be called?                      release-notes-bot
? What should it do?                                     Draft release notes from merged PRs
→ Recommended: Claude Opus 4.8 on Amazon Bedrock · tools: read, web_fetch, web_search · memory on
? Apply this configuration?                              Yes
✓ release-notes-bot is ready — opening chat

You can also skip the wizard and set fields with flags, or start from an existing agent file:

Terminal window
strands --setup                                                     # full interactive wizard
strands --name release-notes-bot --model anthropic/claude-sonnet-5  # set fields directly
strands --agent ./agent.ts                                          # start from an exported agent

Export it to code

Section titled “Export it to code”

When you’re ready to embed the agent in an application, /export writes a Python or TypeScript project with your choices set directly on create_harness(...) / createHarness(...). The project exports a ready-to-import agent, so the CLI is a fast on-ramp to the library below: build interactively, then drop into code.

Use it as a library

Section titled “Use it as a library”

Prefer to write the code yourself? Install the Python or TypeScript package and run your first agent in a few lines.

Install

Section titled “Install”

Requires Python 3.10 or newer.

Terminal window
pip install strands-harness

Run your first agent

Section titled “Run your first agent”

Strands harness runs on the model of your choice and supports model providers across Amazon Bedrock, Anthropic, OpenAI, and Google, plus Ollama for running locally. Amazon Bedrock is the default; pass model="provider/name" to pick another. See choose a model for the full list.

Even a one-line call has a shell, file tools, and web access working out of the box. Strands harness can search the web, compare what it finds, and save the results to a file.

Bedrock is the default, using Claude Opus 4.8 in the region your AWS configuration selects.

from strands_harness import create_harness


agent = create_harness()
agent("Research the three most common strategies for versioning a REST API, compare their tradeoffs, and write a recommendation to api-versioning.md")

Give Strands harness AWS credentials with permission to invoke the model, using one of:

  • Bedrock API key: set AWS_BEARER_TOKEN_BEDROCK to a Bedrock API key. Quickest for local development.
  • AWS credentials: aws configure, or AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and optionally AWS_SESSION_TOKEN as environment variables.
  • IAM roles: on AWS services like EC2, ECS, or Lambda.

Enable access to the models you use in the Amazon Bedrock console; see the AWS documentation.

Keep a conversation across runs

Section titled “Keep a conversation across runs”

Sessions are on by default: Strands harness persists each conversation to disk under ./.agent/sessions with a generated id. Choose the id yourself and a later run rehydrates the same conversation:

from strands_harness import create_harness


agent = create_harness(session={"id": "api-design"})
agent("Which of those would you pick for an API with external customers, and why?")

See persist sessions for how session storage works and how it differs from long-term memory. The CLI persists sessions the same way — resume one with strands --session-id api-design.

The agent you get

Section titled “The agent you get”

However you start it, Strands harness runs on the model of your choice with reasoning on, and it follows a tuned system prompt that tells it to explore before changing things, confirm before anything irreversible, and verify before calling a task done. It also has prompt caching, automatic context management, long-term memory, a generalist subagent, and a todos task tracker enabled by default. See what the default harness does for the full set.

Next steps

Section titled “Next steps”