Claude Think Tool
Implement a dedicated 'think' tool for Claude to perform structured reasoning during complex, multi-step tasks, improving agentic tool use and policy adherence.
When to use
Use when Claude needs to reason through complex problems, make consistent decisions, or follow policies in multi-step agentic tasks, especially when extended thinking isn't sufficient or applicable mid-response.
How to apply
Define a tool named 'think' that takes a single string argument, 'thought'.
Instruct Claude to call this 'think' tool with its reasoning process when it encounters a complex decision point or needs to plan its next action.
Capture the 'thought' output from the tool call.
Do not display the 'thought' to the user; it is for internal reasoning only.
Allow Claude to continue its response generation after the 'think' tool call, incorporating the insights gained.
Glossary
- Tool
- A function or capability that an LLM can call to perform specific actions or access external information, defined by the developer.
- Agentic tool use
- The ability of an LLM to autonomously select and use tools to achieve a goal, often involving planning and decision-making.
- Extended thinking
- An internal Claude capability where the model deeply considers and iterates on its plan before generating a response, distinct from the 'think' tool which can be invoked mid-response.