Set up your workshop environment.
Choose the tools your practice needs, then set up Python and credentials without exposing a key. Most learners need 15–30 minutes.
Requirements and cost at a glance.
| Path | You need | Possible external cost |
|---|---|---|
| Reading + prompt labs | Browser and access to a supported chat model; some product features depend on plan/region | Provider plan, if required |
| Python/API labs | Python 3.10+, Anthropic Console account, API key, prepaid usage | Usage-based; no fixed workshop price |
| Claude Code + SaaS | Git, terminal, Claude Code access; Node/hosting only where the practice says so | Model access and optional hosting/services |
| NanoClaw | Ollama-capable computer; Anthropic API for hybrid/cloud steps; Telegram is optional | Hardware/electricity plus any cloud-model usage |
Compatibility note · reviewed 22 Oct 2026. The examples target Python 3.10+ and pin claude-sonnet-5-5 ($2/$10 per 1M tokens) as the default lab model. Install the current SDKs (anthropic 2.x, mcp 2.x); the Python SDK 2.x sends API version 2026-10-21, which no longer accepts temperature, so the Under the Hood sampling lab is a no-key simulation instead of an API call. Check the release notes and migration guide before a live workshop.
Get an Anthropic API key.
Create a key at the Anthropic Console ↗. It looks like sk-ant-.... API usage requires its own billing setup; do not assume a chat subscription includes API credit.
References: API overview ↗ and current API pricing ↗. Keep the key private. Never commit it to a repo.
Create a virtualenv and run the agent code.
From the repo root, set up an isolated environment, install the dependencies, export your key, then run the day scripts.
cd agents python3 -m venv .venv && source .venv/bin/activate pip install -r requirements.txt export ANTHROPIC_API_KEY="sk-ant-..." python hello_claude.py # Day 01 python tool_use_demo.py # Day 03 python research_agent.py "your question here" # Day 04 capstone
cd agents py -3 -m venv .venv .\.venv\Scripts\Activate.ps1 python -m pip install -r requirements.txt $env:ANTHROPIC_API_KEY="sk-ant-..." python hello_claude.py python tool_use_demo.py python research_agent.py "your question here"
If your organization blocks PowerShell activation, do not weaken the machine-wide policy. Use .\.venv\Scripts\python.exe in place of python, or ask your administrator for the approved workflow.
Run the MCP server and eval harness.
The Agents code lives in workshops/code-examples. With your virtualenv still active and ANTHROPIC_API_KEY exported:
cd workshops/code-examples pip install "mcp[cli]" anthropic python tools_mcp.py # runs an MCP server python eval_harness.py path/to/test_cases.jsonl # eval the agent
Run a local model alongside the cloud.
NanoClaw pairs a local Ollama model with Claude. Install Ollama on the Mac mini (or your machine), then run the triage and router demos.
# On the Mac mini brew install ollama brew services start ollama ollama pull llama3.1:8b-instruct-q4_K_M # Anywhere cd workshops/code-examples pip install feedparser openai anthropic python rss_triage.py # daily local-LLM RSS triage python hybrid.py # cloud/local router demo
That's every runnable script in the library. Other practices are reading + prompts and need no setup beyond a browser at claude.ai.