Monitor Claude Code,
locally.
Wrap claude with onpath and monitor each session you run through it. Detect stuck loops, get a recovery prompt, and let Claude Code inspect local findings over read-only MCP.
Repeated failures
Same test, same error, on repeat — caught by failure signature.
Edit loops
The same path edited repeatedly without a successful command.
No progress & time budget
Configured session budgets crossed without successful progress.
Three steps. Nothing changes about how Claude Code runs.
onpath sits between you and the claude CLI — observing, recording, and reporting without altering the agent's behavior.
Wrap the command
Prefix any Claude Code invocation. onpath spawns it as a child process and watches the stream.
Detect locally
Commands, exits, and edits stream to local SQLite. Detectors flag stuck patterns in real time.
Feed back over MCP
Install the read-only MCP server so Claude Code can read its own findings and recovery prompt.
Claude Code can inspect local evidence — and cannot act through onpath.
The onpath MCP server exposes discovered projects, recent sessions, findings, reports, Work summaries, and deterministic recovery prompts. It has no tool that runs commands, edits files, or changes policy.
Claude Code support at a glance.
onpath wraps the claude CLI on macOS and Linux. Everything runs locally — no agent API keys are shared with onpath.
onpath works with your whole agent stack.
Use the same prefix and the same dashboard across every coding agent on your team.
Put Claude Code on path.
Local-first. Open-core. Two minutes to install and one prefix to start.