HyperProbe: your coding agent debugs prod, no redeploy needed
An SDK and MCP server drop read-only probes into live services, so Cursor, Claude Code, or Codex debug from real variable state instead of logs.
Copy markdown
Virtual breakpoints, straight from your IDE
HyperProbe drops non-breaking probes into running services on demand — no print statements, no redeploy. When live traffic hits the line, it grabs local variables at every frame of the call stack. The company says install takes about 60 seconds and a first investigation around 10 minutes.
How it works: an SDK plus an MCP server
An in-process SDK (Node via the Inspector API, Python via sys.monitoring, Java via JVM bytecode) pairs with an MCP server your agent calls. The agent requests a probe on a suspect line; the SDK captures the whole stack when traffic arrives, at a stated sub-1% CPU overhead.
Read-only by design, PII stripped in-process
Probes read but never write — the founders refused memory mutation so your production never takes a dependency on their infra. Passwords, tokens, SSNs, and card numbers are redacted inside your container before anything is streamed to the agent.
Build this weekend: point an agent at a prod bug
Tell Claude Code or Cursor “checkout returns 200 but orders fail,” and it can inspect the real runtime state at the failure instead of flat logs. It closes the gap where coding agents write code well but debug production blind.