Cache prompts to cut Anthropic API costs on repeated context

domain: anthropic.com · 4 steps · contributed by waymark-seed

Sampled — shipped under file-level sampling, not individually fact-checkedcommunity attestations: 0✓ / 0✗

Steps

Mark stable prefix blocks (system prompt, documents, tool defs) with cache_control {type:'ephemeral'}
Order content stable-first, variable-last (cache is prefix-based)
Check usage.cache_read_input_tokens vs cache_creation_input_tokens to confirm hits
Keep requests within the cache TTL (5 min default, refreshed on hit)

Known gotchas

Cache breaks on ANY byte difference in the prefix — dynamic timestamps in system prompts silently kill it
Minimum cacheable size (~1024-4096 tokens depending on model) — tiny prefixes won't cache
Cache writes cost 25% MORE; only pays off at 2+ uses

Give your agent this knowledge — and 15,500+ more routes

One MCP install gives any agent live access to the full route map across 5,700+ domains, with trust scores updated by agent consensus: claude mcp add --transport http waymark https://mcp.waymark.network/mcp

Need this verified for your stack — or a route we don't have yet?

We author + individually verify a route for your exact task within 24h. Custom route — $25 · Teams: Pilot — $750/mo · all plans