Tool Result

When an LLM emits a tool call, the scaffolding executes it and returns a tool result. In Anthropic's Messages API, tool results are injected as `tool_result` blocks in the conversation. The model receives the content of the result (text, JSON, binary data) alongside the original tool call ID, allowing it to associate results with the specific call that generated them.

Tool results are the primary grounding mechanism in agentic systems. They replace the model's internal knowledge with authoritative external data: a web search result, a database row, a code execution output, or a file read. The quality of downstream reasoning depends heavily on how faithfully the tool result represents ground truth.

Error handling in tool results is a critical design decision. Returning a structured error (`{"error": "rate_limited", "retry_after": 30}`) is far more useful than returning an empty response or raising an exception, because it gives the model actionable information to decide its next step—wait and retry, try a different tool, or escalate to a human.

## Related Terms

- [[glossary/tool-use|Tool use]] — the protocol Claude follows when invoking external tools from an agent loop - [[glossary/function-calling|Function calling]] — the mechanism by which the model emits structured JSON to invoke an external function - [[glossary/parallel-tool-calls|Parallel Tool Calls]] — the capability to invoke multiple tools simultaneously in a single model turn - [[glossary/agent-loop|Agent Loop]] — the iterative perceive-act-observe cycle the harness executes - [[courses/mcp-from-first-principles-to-production|Course: MCP from First Principles to Production]] — hands-on practice with the concepts covered in this entry