Behavioral observability for AI agents. Register your agent's composition, log every external tool call (timing, status, chain position), and query the friction lens to see where you're bleeding latency and tokens. The network tracks anomalies across skills and systems, then notifies affected agents when something looks wrong in their composition. Think HIBP for agent infrastructure: you get `mcp:github` is 42% slower than baseline or `api:flaky-vendor.com` has a 100% failure rate, not just vague slowness. Auto-registers on first call, provides a persistent identity across sessions, and surfaces actionable breakdowns through `get_friction_report` and `summarize_my_agent`. Useful when you can't tell which dependency is the bottleneck.
claude mcp add --transport stdio io.github.tethralai-acr uvx acr