lastping-app

Purpose

LastPing is used to monitor AI agents, cron jobs, and CI/CD pipelines. Its monitoring logic is the reverse of conventional approaches: most monitoring actively checks the target to see whether it has become abnormal; LastPing instead waits for the target to "check in" on its own and alerts when the target does not check in. This inversion is the core of the product—a failed task cannot send an error, but it can "send nothing," and "absence" is precisely the only signal that a failed process can still produce.

Main Capabilities

• Alerting scenarios: send notifications when an AI agent fails, stalls, waits for user approval, or never starts.
• Environment monitoring: simultaneously observe the cron jobs and CI/CD pipelines around the agent.
• Client compatibility: supports Claude Code, Codex, Cursor, and any agent connected via MCP.
• Command-line wrapping:lastping run can wrap any command and report startup, exit code, and trace.
• Exit code passthrough: the wrapper exits with the exit code of the wrapped command, so CI behavior is the same as before adding it.
• Does not affect the original command: when LastPing is unreachable, the task runs, outputs, and exits normally as usual.
• Preserves interactivity: stdin and stdout are handed over as file descriptors, so it can wrap REPL or agent sessions.
• Automatically configures OpenTelemetry: automatically sets OTLP export-related variables in the wrapped command's environment, so auto-instrumented agents can export trace spans without code changes.
• MCP server: agents can create monitors themselves, retrieve ping instructions, and integrate them into their own workflows.
• Terraform Provider: manage as "monitoring as code."

Usage

• Command line: via lastping run wrap an existing command; precompiled binaries can be installed without the Go toolchain (macOS, Linux, amd64/arm64), and the Windows version is available on the releases page.
• MCP connection: the host server is the recommended path, at the address https://mcp.lastping.dev/mcp. Claude Code can connect directly; Cursor and Windsurf use a URL plus API key; Claude Desktop can use a one-click extension (lastping.mcpb, no Node.js required) or mcp-remote bridging, and custom connectors are also supported.
• ping mechanism: each monitor has a URL, and anything that can make an HTTP request can report; you can add ?rid=<id> to pair the start and result of a run.
• Trace reporting:POST https://ping.lastping.dev/v1/traces accepts OTLP/HTTP export, supporting Bearer key or headless exporter methods; spans require the resource attribute lastping.monitor_id, and with lastping.run_id they can be grouped into the same run; the payload limit is 1 MiB after decompression, 500 spans per request, and 2000 spans per run.
• Toolset: MCP provides tools for monitor management, discovery, reporting, events and runs, tracing, agent observability, failure loops, alert routing, delivery logs, targets, alert templates, agent registry, status pages, API keys, Terraform export, and more. Among them get_ping_instructions Allows the agent to create monitoring, obtain the ping command, and get it working in a single conversation.

Use cases

Suitable for developers and teams that need to ensure AI agents, scheduled tasks, and CI/CD pipelines run reliably. Typical scenarios include: being notified promptly when an agent fails or gets stuck, a cron job forgetting to report, a CI pipeline being interrupted, and needing to manage monitoring as infrastructure as code. Free for personal use.

Limitations

• Windows builds are on the releases page and are not provided through the installation script.
• When Claude Desktop is bridged through mcp-remote, Node.js is required, and the full path to npx must be provided.
• One-click extension uses a self-signed certificate, and Claude Desktop will show it as unverified.
• Trace payloads have size and span count limits.
• Other platforms and pricing information: not yet provided.