← Back to all sparks
O

Oh Dear

INFRA · APIS
Velocity6.3

Website monitoring service covering uptime, broken links, certificates and scheduled tasks

Oh Dear turns its entire monitoring surface into an MCP API, making the product fully AI-agent accessible.

uptime-monitoringmcpai-agentsdevtoolsapisite-reliability
Current state
Oh Dear is an uptime and site monitoring product that has spent the last month systematically wiring its full feature set into a Model Context Protocol server. The August 28 release added MCP tools for downtime data, check summaries, domain intelligence, maintenance windows, check state, status page management, notification destinations, scheduled tasks, tags, Broken Links, Application Health, and account management—covering virtually the entire product in one release. The following week, API token scoping was hardened specifically to secure MCP tool access.
Where it's heading
Oh Dear is deliberately building an AI-first monitoring interface, not as a side project but as a primary access layer. The sequence—ship comprehensive MCP coverage, immediately harden authorization around it—suggests they're treating the MCP server with the same production seriousness as their REST API. The internal migration from Prism to Laravel AI SDK for their AI monitoring features and the addition of automated prompt evaluations show the same investment on the engineering side. JSON request body support in HTTP checks and Arazzo specification updates round out a broader API-maturity push.
Prediction
Expect Oh Dear to publish MCP server documentation or a dedicated landing page as the natural next step given coverage breadth. AI-generated downtime summaries or weekly digest features inside the product UI are a logical extension of the agent-accessible data layer they've now built.

Recent moves

  1. 11d ago

    Changelog: API docs expanded with response body assertions and Arazzo workflow

    Expanded OpenAPI spec to include response body assertions and HTTP uptime settings previously missing from documentation, and updated the Arazzo specification with a full monitoring workflow example. A documentation and developer-experience improvement that makes Oh Dear's API surface easier to integrate against programmatically.

    View source ↗
  2. 18d ago

    Changelog: 2026-08-28

    ⚡ SPARK

    A single release adds MCP tools covering downtime reports, check summaries, domain intelligence, maintenance windows, check state management, status page updates, notification destinations, scheduled tasks, tags, Broken Links, Application Health, and account management. Oh Dear's full product is now addressable by AI agents in one sweep.

    View source ↗
  3. 25d ago

    Changelog: API token scoping and MCP authorization hardened

    API token scoping was strengthened so restricted tokens only access permitted monitors, and MCP tool authorization was tightened to require appropriate permissions for write actions. Security hardening that followed directly from the MCP expansion—good practice given the broad new attack surface.

    View source ↗
  4. 1mo ago

    Changelog: 2026-08-14

    Two bug fixes: recovery notifications were missing after performance delta alerts, and monitors on metered billing could be incorrectly deleted by the monitor limit check. Both affect edge cases in billing and alerting logic.

    View source ↗
  5. 1mo ago

    Changelog: HTTP checks get raw JSON request bodies; AI monitoring migrated to Laravel AI SDK

    HTTP uptime checks now accept raw JSON request bodies with syntax highlighting, enabling monitoring of JSON API endpoints without workarounds. Oh Dear also migrated its internal AI monitoring from Prism to the Laravel AI SDK and added prompt evaluation automation—internal infrastructure improvement that doesn't change the user-facing feature set.

    View source ↗
  6. 1mo ago

    Changelog: 2026-07-31

    Application Health metadata item limit raised from 20 to 40 and marketing attribution was improved. The Claude skill addition is an internal engineering tool, not a user-facing feature. Routine improvements with no significant direction signal.

    View source ↗