MCP Server

What devhelmhq/mcp-server does

devhelmhq/mcp-server exposes DevHelm monitoring and alerting operations through the Model Context Protocol. Agents can work with uptime monitors, incidents, alert channels, and notification policies without switching to a separate DevHelm client.

Monitor operations cover the full lifecycle: listing and inspecting monitors, creating configurations for supported probe types, editing or deleting them, pausing and resuming checks, and triggering an immediate test. Recent check results and monitor version history are also available.

Incident tools support workspace-wide listing, detailed retrieval, manual creation, resolution, and timeline analysis. A timeline includes state transitions, the rule evaluations behind triggering transitions, and the policy snapshot active at the time.

How it works

The server can run as a hosted MCP endpoint at https://mcp.devhelm.io/mcp or locally over stdio. Hosted connections authenticate with a bearer token or, for clients that only accept a URL, an API key embedded in the endpoint path. Local execution uses the devhelm-mcp-server command or uvx.

Detection-focused tools provide historical context rather than only current status. A check trace ties one check execution ID to its rule evaluations, resulting transitions, and active policy snapshot. Policy snapshots are retrieved by content-addressed SHA-256 hash, allowing historical evaluations to be examined against the exact policy that was active at the time. Monitor rule evaluations and state transitions can be paginated and filtered by rule type, region, match status, or ISO-8601 time bounds.

Setup and configuration

For local use, install the package with pip, set the API token, and start the server:

pip install devhelm-mcp-server
export DEVHELM_API_TOKEN=your-token
devhelm-mcp-server

The uvx alternative avoids a separate package installation:

export DEVHELM_API_TOKEN=your-token
uvx devhelm-mcp-server

Hosted clients can use either https://mcp.devhelm.io/mcp with an Authorization: Bearer <your-api-token> header or https://mcp.devhelm.io/<your-api-token>/mcp when only a URL can be configured. Cursor and Claude Desktop can be configured with the hosted URL in their MCP server settings; Windsurf is also named as a supported MCP client.

Tools and capabilities

• Create monitors for HTTP, DNS, TCP, ICMP, heartbeat, MCP server, browser, and multi-step API checks.
• Set monitor frequencies from 10 to 86,400 seconds, subject to the plan minimum.
• Pause, resume, test, update, and delete monitors, then inspect results and version history.
• Create and resolve incidents with severity values of DOWN, DEGRADED, or MAINTENANCE.
• Reconstruct detection history through incident timelines, check traces, rule evaluations, and state transitions.
• Manage alert channels such as email, webhooks, Slack, PagerDuty, Opsgenie, Teams, Discord, and other listed destinations.
• Define notification policies with match rules, escalation steps, enablement state, and evaluation priority.

Monitor configuration is required for probe-based monitor types, while browser and multi-step monitors can use their type default when configuration is omitted. Monitor attribution is assigned server-side as MCP and cannot be overridden through these tools. Alert channel configuration is type-specific, and notification policies require match rules, escalation steps, an enabled flag, and a priority.

Limitations and notes

Creating a monitor requires a name and type; probe types also require configuration. Creating an incident requires a title and severity, with monitor association and a detailed body available optionally. Detection history and result collections use pagination where specified, so callers may need to follow cursors for complete results. Deleting monitors and alert channels is permanent, while pausing a monitor stops checks until it is resumed.