healthchecks-mcp

healthchecks-mcpCron monitoring, in the conversation

MCP server for Healthchecks — inspect, create and adjust cron and uptime checks, and read why one failed

Get started
Tools reference
GitHub

  • Answers why it failed

    get_ping_body returns the output a job POSTed when it reported failure — the stack trace, not just the red dot. list_flips gives the history behind the current status, narrowed to a time window.

  • Read-only keys work properly

    Healthchecks hands a read-only key a different object — a unique_key where a uuid would be, and no ping URLs at all. Checks are addressable by either, and get_api_key_info reports up front which kind of key is configured and which tools it cannot reach.

  • Only the tools you want

    HEALTHCHECKS_READ_ONLY=true registers the read tools and nothing else. HEALTHCHECKS_ALLOW_TOOLS cuts finer — essential for a curated handful, your own comma-separated list, or a whole family with list_* — and HEALTHCHECKS_DENY_TOOLS subtracts. Whatever is filtered out does not exist on the protocol rather than failing when called, and a name that matches no tool stops the server at startup instead of quietly going missing.

  • Safe by default

    delete_check asks a person first, through MCP elicitation — a dialog the model cannot answer on its behalf, falling back to a single-use token where the client cannot show one. pause_check deliberately does not ask: resume_check puts it back. Upstream content is marked untrusted, and read-only mode simply does not register the write tools.

How healthchecks-mcp connects an MCP client to HealthchecksAn MCP client talks to healthchecks-mcp over stdio. The server calls the Healthchecks Management API v3 over HTTPS. The monitored jobs ping Healthchecks directly, on a path the server is not on.MCP clientClaude, Codex, …healthchecks-mcp14 tools, 7 essentialasks a personHealthcheckscron and uptimestdioHTTPS/api/v3Your jobscron, timers, backupsping
The server only reads and writes checks. The pings that keep them green come from the jobs themselves, on a path this server is deliberately not on.

Fourteen tools, or seven ​

The tool list is not fixed. HEALTHCHECKS_ALLOW_TOOLS=essential narrows it to the seven that cover the job end to end, and a name that matches nothing stops the server instead of quietly leaving a tool out of tools/list.

Listing the tools, narrowing them to the essential preset, and the startup abort a mistyped tool name produces

Recorded from docs/demo.tape with no credentials at all — the server starts without an API key so that registries and inspectors can introspect it.

Running it elsewhere ​

A client that cannot spawn a local process — ChatGPT connectors, Claude on the web, Cursor, LibreChat — cannot start healthchecks-mcp the way Claude Code does. mcp-hub is the bridge: one container serves many stdio MCP servers over Streamable HTTP, with an OAuth 2.1 login behind a single password and long-lived tokens for the clients that cannot do OAuth. Its /hub endpoint puts every server behind six meta-tools, so one connector reaches all of them without N×tool schemas in the model's context, and it speaks both protocol revisions — a question this server asks travels through it to the person at the far end instead of ending at the gateway.

Its configuration is Claude Code's mcpServers format, so the entry you already have is the entry it takes: Through mcp-hub.

Released under the MIT License.

Copyright © 2026 Willi Thiel