MCP Server

Drive Enori from Claude Desktop, Claude.ai custom connectors, or any client that speaks the Model Context Protocol. 45 tools — monitors, incidents, alerts, status pages, browser checks, and SLOs — all from natural language.

What you can do

Once connected, ask Claude things like:

• "Which of my monitors had incidents in the last 24 hours?"
• "Pause the API monitor for an hour while I deploy."
• "Create a 99.9% availability SLO with a rolling 30-day window on monitor X."
• "What's the current burn rate on my customer SLA?"
• "Generate a status page for the public-facing services."

Claude routes the question through the relevant Enori MCP tools, fetches your data, and answers in plain language. No copy-pasting between dashboards.

Connecting

The MCP server lives at https://mcp.enori.io/mcp. There are two ways to authenticate:

OAuth (recommended for Claude Desktop / Claude.ai)

OAuth gives the client full access to your account, just like a browser session. No scopes to manage, no API keys to rotate.

Claude Desktop / Claude Code CLI:

{
  "mcpServers": {
    "enori": {
      "url": "https://mcp.enori.io/mcp"
    }
  }
}

The first time Claude calls a tool, your browser opens to https://app.enori.io for sign-in. After approval, the client stores a refresh token (30-day TTL, rotated on each use) and the dance is invisible from then on.

Claude.ai custom connectors:

In Claude.ai, Settings → Custom Connectors → Add custom connector → URL = https://mcp.enori.io/mcp/sse. The legacy SSE transport is on the /mcp/sse path; modern HTTP-streaming clients use /mcp. Both paths talk to the same server.

API key (recommended for scripts / CI)

API keys are scope-bound. The MCP server reuses the same scopes as the REST API:

Generate keys in Settings → API Keys in the dashboard. Each key can hold multiple scopes; pick the smallest set the use case needs.

Config:

{
  "mcpServers": {
    "enori": {
      "url": "https://mcp.enori.io/mcp",
      "headers": { "X-Api-Key": "YOUR_KEY" }
    }
  }
}

Rate limits

Hitting the limit returns a 429 with a Retry-After header. Bursts well below this on natural conversational use; the limit is mostly there to catch runaway scripts.

Tool catalog

Tools are grouped by what you'd reach for them to do. Every read tool requires a read scope; every write tool requires the corresponding write scope. Browser-monitor and SLO tools share the regular monitors: scopes — there is no separate browser: or slos:* scope.

Monitors

Incidents

Alerts & status

Maintenance windows

Browser monitors

Browser monitors run real Playwright sessions through the warm-pool executor. The MCP surface here mirrors the dashboard's browser tab.

Traces (Website monitors)

Job monitors

SLOs

See the Service Level Objectives guide for the concepts behind these tools — windows, burn rate, the three SLI types, exclusions, and the SLA-report flow.

Examples

Smoke-check uptime over coffee

You: Anything degraded right now? Claude: Calling get_health_summary… two monitors are burning. The API monitor is at 34 % error rate over the last hour (last successful check 12 minutes ago, response time spiking from 180 ms to 4.2 s); the staging environment is showing intermittent SSL errors (3 of the last 10 checks failed handshake). Other 18 monitors green. Want me to acknowledge an incident on the API one or check its recent traces?

Set up a customer-grade SLA from scratch

You: Set up a 99.5 % monthly availability SLA on the production API monitor with a customer-friendly description. Claude: Calling list_monitors to find the production API monitor… found it (mon_abc123). Calling create_slo with availability SLI, calendar-monthly window, target 99.5, description "MSA §4.2 — monthly uptime commitment for paying customers", maintenance + paused exclusions on, balanced burn-rate alerts (fast + slow). Created slo_def456. The first monthly window starts 2026-05-01 00:00 UTC and the achieved % so far is 100 % (no incidents yet this month).

Investigate a slow check

You: Why was check chk_xyz on monitor X slow? Claude: Calling get_check_results to find that check (response time 8.3 s vs typical 220 ms)… it has trace data. Calling get_trace_detail. The check spent 7.6 s in a single Postgres query (SELECT … FROM users WHERE …) inside the /api/users endpoint — that's where the budget went. Database span has db.statement truncated to 200 chars; it's a query against a 50M-row table without an index hint. Want me to open an incident with this evidence pre-filled?

Errors

The MCP server returns errors as a normal JSON tool result with an error field rather than HTTP 5xx — this is what most MCP clients render correctly. Common shapes:

For OAuth, refresh tokens rotate on every use. If you let a token sit unused for 30 days the next call returns Refresh token expired — the client will re-prompt you to sign in.

Self-host the internal flavor (advanced)

The public server at mcp.enori.io is the recommended path. There's also an internal stdio version (tools/enori-mcp-server in the repo) with extra debugging and admin tools. It's intended for developers building against Enori — not for production agents — and runs locally over stdio rather than HTTP. Set ENORI_API_KEY (and optionally ENORI_ADMIN_KEY for admin tools) in your shell env, then add it to .claude/mcp.json:

{
  "mcpServers": {
    "enori-internal": {
      "command": "node",
      "args": ["./tools/enori-mcp-server/dist/index.js"],
      "env": {
        "ENORI_API_KEY": "YOUR_KEY",
        "ENORI_API_URL": "https://api.enori.io"
      }
    }
  }
}

The internal server adds manage_* consolidated CRUD tools, admin observability, and direct DB validation helpers. See the repo for full surface details.

Status

The public MCP server has been live at mcp.enori.io since 2026-04. Tool count grows when the dashboard surface grows — recently added: SLO management (May 2026), browser monitor variables (April 2026), Job-monitor custom metrics (April 2026). The contract is additive; we don't break existing tools.

Spotted a missing tool, hit a confusing error, or want a workflow that needs a new tool? Email support@enori.io.