agentic_ai_basics

tool_use

Tool use (function calling) — giving Claude the ability to take actions, not just talk.

basic_agentic_tools.py

A task manager agent that uses tools to add, list, and complete tasks. Tasks persist to tasks.json in this folder. Type exit to end the conversation.

Concepts covered

• JSON Schema tool definitions — each tool declares its inputs (required/optional params, enums) so Claude knows how to call it.
• tool_choice — defaults to "auto", letting Claude decide when a tool is needed. Alternatives: force a specific tool, force any tool, or disable tools entirely.
• Parallel tool calls — if a request needs multiple tools at once (e.g. “add a task and show me my list”), all calls are executed and their results returned together in a single message.
• Tool result handling, including errors — a failed tool call (e.g. completing a task ID that doesn’t exist) is returned with is_error: true so Claude can react appropriately instead of the program crashing.
• The manual agentic loop — two nested loops: an outer loop (in main()) that keeps the conversation going across turns, and an inner loop (in run_turn()) that resolves any tool calls within a single turn before returning.
• Stateful tools — tools have real side effects (reading/writing tasks.json), not just pure functions.

Run

From the repo root:

pip install -r requirements.txt
export ANTHROPIC_API_KEY=your-key-here
python3 Core_Architecture/tool_use/basic_agentic_tools.py

Try prompts like:

You: Add a task to buy groceries, due tomorrow
You: What tasks do I have open?
You: Mark that task as done

Configuration

Edit the constants at the top of basic_agentic_tools.py:

• MODEL — the Claude model used (default: claude-sonnet-5)
• MAX_TOKENS — max response length
• EFFORT — thinking/response depth (default: medium)
• SYSTEM_PROMPT — instructions that shape Claude’s behavior as the assistant
• TOOLS — the JSON Schema tool definitions Claude can call
• TASKS_FILE — where tasks are persisted (default: tasks.json next to the script)

See also

• ../../Execution_Loops/agentic_loop/basic_agentic_loop.py — the outer chat loop this template’s main() follows
• ../../Agent_Frameworks_and_Patterns/tool_registry/basic_tool_registry.py — a scalable alternative to this file’s hand-maintained TOOLS list + if/elif dispatch
• ../agent/README.md — this same tool-calling loop shape, repurposed for unattended goal pursuit instead of turn-by-turn chat
• ../../Tools_and_Actions/README.md — server-side and Anthropic-defined client-side tools, building on this custom client-side pattern