Monitor Your Stack with Claude and the Statusfield MCP Server

When an incident starts, the first question is almost always the same: is this us, or is it them?

Checking takes thirty seconds. You alt-tab, load a status page, squint at the last-updated timestamp, decide it's probably fine, alt-tab back. Then you do it for the next service on your list. By the time you've ruled out GitHub, Stripe, and Vercel, you've lost the thread of whatever you were debugging.

The Statusfield MCP server cuts that loop out entirely. It connects Claude to the thousands of vendor status pages we aggregate, so you can ask "is anything down?" without leaving your editor or your terminal.

This is a setup guide. The announcement post covers the why. This one covers the how — Claude Code, Claude Desktop, and a few workflows worth building.

What the MCP server exposes

Four read-only tools. All of them speak the same sf_ API key:

The important distinction: list_my_down_services reads your workspace's monitor list. The other three read the shared catalog. If you want "is anything I depend on down?", that's the first tool. If you want "is Stripe down, period?", that's get_service_status.

Both are useful. Most incidents start with the first question, escalate to the second.

Setup: Claude Code

One command:

claude mcp add --transport http statusfield https://statusfield.com/api/mcp \
  --header "Authorization: Bearer sf_your_key_here"

Replace sf_your_key_here with a key from Settings → API Keys in your Statusfield dashboard. The key is shown once when you create it — copy it then.

Verify the connection:

/mcp

You should see statusfield listed as connected. If it's not, check that the key starts with sf_ and that you're on the Hobby plan or above — the API and MCP are paid features.

Once it's connected, Claude Code can answer status questions in any conversation. You don't need to invoke it explicitly — Claude will call the right tool when the question makes sense.

Setup: Claude Desktop

In your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on Mac):

{
  "mcpServers": {
    "statusfield": {
      "command": "curl",
      "args": [
        "--mcp",
        "--header", "Authorization: Bearer sf_your_key_here",
        "https://statusfield.com/api/mcp"
      ]
    }
  }
}

Restart Claude Desktop after saving. The Statusfield tools will show up in the tool picker.

Workflows worth setting up

Incident triage

The most common use. You're mid-incident and you want to rule out vendor causes before you start reverting your own code.

You: We're seeing 500s on checkout. Is anything I monitor down?

Claude: (calls list_my_down_services) Stripe is showing degraded — specifically
the Payment Processing component. Everything else you monitor is operational.
GitHub, Vercel, Twilio are all fine.

Your 500s line up with the Stripe incident. This is probably not your code.

Claude calls the tool, formats the result, and gives you a conclusion — not a list of links to go check. The thirty-second tab-switching loop becomes one message.

Pre-deploy context

Before you trigger a deploy from Claude Code, ask first:

You: Before I push this auth refactor — anything in my stack down right now?

Claude: (calls list_my_down_services) All clear. Your monitored services are
all operational. Safe to deploy.

This is the lightweight version of the full pre-deploy API check. No shell script, no CI step — just a question before you push. It doesn't gate the deploy automatically, but it catches the obvious cases before you waste a deploy cycle.

Catalog lookup

You're adding a new dependency and want to know if Statusfield tracks it before you monitor it:

You: Do you know if Statusfield tracks Resend's status?

Claude: (calls search_services for "Resend") Yes — Statusfield monitors Resend.
You can add it to your monitor list from the dashboard, or check its current
status directly.

If it's in the catalog, you get a confirmation. If it's not, Claude says so — and that's useful information too (it means you'd be relying on a service that's harder to monitor at scale).

Morning context

Some people start the day with a status sweep:

You: Quick rundown before standup — anything in my stack showing issues?

Claude: (calls list_my_down_services) One flag: PagerDuty is showing partial
degradation on the Webhooks component — started about 40 minutes ago. Everything
else is clean: GitHub, Stripe, Vercel, AWS, Datadog all operational.

Worth mentioning in standup if your team uses PagerDuty for alerting.

This is the "daily briefing" version — useful if your team depends on a lot of services and you want a single status check before the day starts.

What the MCP server doesn't do

A few things worth being clear about:

It doesn't beat the vendor. We relay the official vendor status — we don't detect outages before the vendor reports them. When Stripe flips a component to degraded, we mirror that, fast and faithfully. The value is having thousands of those streams consolidated and queryable from a single tool.

It's read-only. Claude can't create monitors, delete them, or change your account through these tools. That's intentional — a read-only interface can't misfire. Write capabilities are on the roadmap.

It uses your monitor list for list_my_down_services. If you haven't added a service to your monitor list in Statusfield, that tool won't tell you about it. The catalog tools (get_service_status, list_outages) cover everything we track regardless of your monitor list — but scoped queries require scoped setup.

On context in agent loops

If you're running Claude Code as an autonomous agent — letting it make commits, open PRs, run test suites — the status tools become a natural part of the context. An agent can check whether its dependencies are up before it starts a long-running task:

Agent reasoning: I'm about to run the full test suite and open a PR. 
Let me check whether anything I depend on is currently degraded.
→ calls list_my_down_services
→ Stripe degraded, Webhooks component
→ Test suite includes payment tests. Flagging before running.

The agent doesn't need a separate integration for this — it's the same MCP call, same tools, same result. Status context becomes available to any agent that has the Statusfield tools connected.

Getting started

You need a Statusfield account on the Hobby plan or above for MCP access. The free plan doesn't include it.

Steps:

1. Sign up or upgrade if you're not already on Hobby+
2. Add the services your team depends on to your monitor list
3. Create an API key at Settings → API Keys
4. Run claude mcp add with the command above (or update claude_desktop_config.json for Desktop)

The MCP server is at https://statusfield.com/api/mcp and documented at statusfield.com/docs/mcp. The REST API reference is at statusfield.com/docs/api if you want to build the same checks into scripts or CI.

Questions or feedback on the MCP server? support@statusfield.com — I read all of these.