Clean Up Agent Context Using Web Search Parameters
Optimize agent memory by dropping redundant web search blocks.
Implement the `response_inclusion` parameter in Claude’s web search tools to automatically remove unnecessary data from API responses.
The Scenario
You are building a research agent that performs multiple consecutive web searches. You want to keep the conversation history clean so the agent stays focused on the findings rather than re-reading the raw HTML results.
Before & after
Developers previously had to write custom scripts to parse and 'clean' the search result blocks out of the AI's response history to prevent context clutter. This took 20–30 minutes of coding and testing.
By passing the `response_inclusion` parameter in the Web Search/Fetch tools, the AI drops redundant data from the API response. This streamlines the agent's logic in 2–3 minutes.
The Prompt
// Use the new web search tool parameter to optimize agent context
{
"tools": [{
"name": "web_search_20260318",
"parameters": {
"query": "[SEARCH_QUERY]",
"response_inclusion": "drop_consumed_blocks"
}
}]
}The latest Web Search and Web Fetch tools (`web_search_20260318`) allow for more efficient agentic workflows by dropping consumed result blocks. This prevents the context window from becoming bloated with raw search data that the agent has already processed.
Source
Claude Platform - Claude API Docs"adding a response_inclusion parameter to drop consumed result blocks from the API response for agentic workflows."