Step 3 — Web Search

For agents: Verify web search is available, run a live test query, and confirm you can retrieve current information before moving to Step 4.

A Claw without web search only knows what it was trained on. That's useful but limited — no current news, no live prices, no recent docs. Web search turns your Claw into a research partner that can look things up on demand.

This step verifies that web search is enabled and working. The morning brief in Step 6 depends on it.

Check if web search is already available

Send your Claw this message:

Search the web for today's top headline and tell me what you find.
Include the source URL.

If your Claw returns a current result with a URL, web search is already working. Skip to the verification step.

If your Claw says it can't browse the web, continue to the next step.

Enable web search via skill

Create a web search skill file in your workspace:

cat > ~/skills/web-search.md << 'EOF'
---
name: web-search
description: Search the web for current information using available search tools
argument-hint: "<search query>"
---
 
Search the web for: $ARGUMENTS
 
Use the best available search tool (Brave, Bing, DuckDuckGo, or any configured search MCP).
 
Return:
- The top 3–5 relevant results
- A brief summary of what you found
- Source URLs for each result
 
Always include the date of the information when it is time-sensitive.
EOF
mkdir -p ~/skills
mv ~/skills/web-search.md ~/skills/web-search.md 2>/dev/null || true

If your HyperClaw environment has a search MCP configured (Brave Search, Bing, etc.), the skill will route to it automatically.

Configure a search provider if needed

Check which search tools are available:

# Check for configured MCPs
ls ~/.openclaw/mcp/ 2>/dev/null || echo "No MCP directory found"
cat ~/.openclaw/config.json 2>/dev/null | python3 -c "import sys,json; d=json.load(sys.stdin); print(json.dumps(d.get('tools',{}), indent=2))" 2>/dev/null

If no search tool is configured, add one. Brave Search is the recommended option:

# Add Brave Search MCP (requires a free Brave Search API key)
# Get your key at: https://brave.com/search/api/
cat >> ~/.openclaw/config.json << 'EOF'
{
  "tools": {
    "brave-search": {
      "type": "mcp",
      "command": "npx @modelcontextprotocol/server-brave-search",
      "env": {
        "BRAVE_API_KEY": "your_key_here"
      }
    }
  }
}
EOF

Restart your Claw after adding the tool.

Verify with a live query

Once search is enabled, send this test:

Search the web for "HyperClaw agent platform" and summarize what you find.
Include at least one source URL.

A working search returns a real result with a URL. The URL should be something you can open and verify.

If your Claw is running in an environment with an existing search MCP already configured (common in HyperClaw pods), you may find search just works after enabling the skill file.

How your Claw uses search

After setup, web search happens naturally. Your Claw will search when it needs current information and when you ask it to:

What's the weather in New York this morning?
Search for the latest Claude model release notes.
Find me three articles about Kubernetes pod autoscaling from the last month.

The morning brief in Step 6 uses web search for news and weather — which is why this step comes first.

✅

Verify before continuing: Ask "Search the web for today's date and tell me what day it is." If your Claw returns today's correct date from a live search result (not from training), web search is working.

Next: Step 4 — Memory

2 — ZenBin 4 — Memory