Distributing tools and configuring tool_choice
Why each agent should hold only its role's tools, how to scope them with tools, disallowedTools and inline mcpServers (and why allowedTools only auto-approves), and every tool_choice value: auto, any, a forced tool and none, including forcing a first step and the models that reject forced choice.
Distribute tools appropriately across agents and configure tool choice
~8 min read · 14 practice questions ·
What the exam expects
- Explain why too many tools (18 instead of 4-5) or tools outside an agent's specialization degrade selection, and redesign around role-scoped tool sets.
- Give a specialist one narrow cross-role tool for a high-frequency need while routing complex cases through the coordinator, and replace generic tools with constrained ones.
- Restrict a subagent's tools with
tools,disallowedToolsor inlinemcpServers, knowing thatallowedToolsonly auto-approves. - Pick the
tool_choicevalue for a requirement:auto,any, a forced{"type": "tool", "name": ...}ornone. - Force a first step, then return to
autofor follow-up turns, and know where forced choice is rejected.
Key terms: scoped tool access (each agent gets only the tools its role needs); cross-role tool (a narrow tool from another role's domain, such as verify_fact); tool_choice (the request field controlling whether and which tool Claude must call); forced tool use (any or a named tool); auto-approval (a permission rule that skips the prompt without adding or removing tools).
Key facts
tool_choice- Detail
autowhen tools are provided;nonewhen no tools are provided
any vs tool- Detail
anyrequires some tool;toolrequires the named tool
- Detail
- None: the API prefills the assistant message for
anyandtool
- Detail
any/toolfail on some models and with manual extended thinking;autoandnonestill work
- Detail
strict: true, nottool_choice
- Detail
disable_parallel_tool_use: trueinsidetool_choice
tools field- Detail
- Inherits every tool available to subagents
- Detail
disallowedTools: mcp__serverormcp__server__*
allowed_tools- Detail
- Auto-approves; unlisted tools remain available
- Detail
- 4-5 rather than 18 (exam-guide guidance)
Sources for this section (6)
- Define toolsplatform.claude.com (opens in a new tab)
- Strict tool useplatform.claude.com (opens in a new tab)
- Parallel tool useplatform.claude.com (opens in a new tab)
- SDK subagentscode.claude.com (opens in a new tab)
- Configure permissionscode.claude.com (opens in a new tab)
- Exam guideeverpath-course-content.s3-accelerate.amazonaws.com (opens in a new tab)
- Why does giving every subagent all 16 tools hurt reliability?
- When should you use
tool_choice: anyinstead of a named tool? - Which field keeps a subagent's inherited tools but withholds one MCP server?
- Why might a request with tools and no
tool_choicereturn only text?
The questions are free to read. Answers with their sources and self-grading are included with the pack.
Keep reading with the Claude Certified Architect - Foundations pack
- Core concepts
- Exam traps
- Worked scenario
The pack includes:
- All 30 lessons of the Claude Certified Architect - Foundations guide, with reading progress
- Interactive decision walks and step-by-step sequence steppers
- Inline practice questions and one-click topic practice (14 questions for this lesson)
- An official source cited for every claim
- An AI tutor that answers from the guide
- Unlimited timed practice exams
Practice for the Claude Certified Architect - Foundations exam
Certifyforce is an independent study resource and is not affiliated with, endorsed by or sponsored by Anthropic. Claude is a trademark of Anthropic.