cc-web

How often does Claude Code use web search? Shows WebSearch vs WebFetch split, session adoption rate, and per-project breakdown.

cc-web — Claude Code web search & fetch usage

  Total sessions:       755
  Sessions w/ web:      96 (12.7% of sessions)
  Total web calls:      2,460
  WebSearch:            1,421  (57.8%)
  WebFetch:             1,039  (42.2%)
  Avg per web session:  25.6 calls
  Peak in one session:  794

────────────────────────────────────────────────────────
  Web calls per session (sessions with web)

  1-2      ██████████░░░░░░░░░░░░░░    17  (17.7%)
  3-10     ████████████████████████    40  (41.7%)
  11-50    ███████████████████░░░░░    31  (32.3%)
  51+      █████░░░░░░░░░░░░░░░░░░░     8  (8.3%)

Usage

npx cc-web          # Web search and fetch stats
npx cc-web --json   # JSON output

What it shows

• Adoption rate — percentage of sessions that used web search or fetch
• WebSearch vs WebFetch — split between search queries and direct URL fetches
• Distribution — single web call vs research-heavy sessions
• Peak session — the session with the most web activity
• By project — which projects rely on external knowledge most

What the data means

87% of Claude Code sessions never touch the web. Most work is purely local — reading files, editing code, running Bash. When web access does happen, it tends to cluster: the average web-using session makes 25.6 calls, and the peak session made 794.

• WebSearch (57.8%): query-based searches for documentation, APIs, best practices
• WebFetch (42.2%): direct URL fetches — often following up search results

Privacy

Reads session file content to count tool call occurrences. No content is transmitted. Everything runs locally.

Browser version

Drop your ~/.claude folder into cc-web on the web — no install required.

Part of cc-toolkit — 60 free tools for Claude Code