MCP server
Ask Claude, ChatGPT, Cursor or any MCP client about your monitors, and let it act on them: run checks, acknowledge incidents, schedule maintenance and post status updates.
Connect with OAuth
Clients that support MCP authorization only need the server URL. They register themselves, send you to Observe to sign in, and you choose the organization, read only or read and write, and optionally which monitors before you click Allow. No token to copy; disconnect any time in Profile → API & data.
Server URL: https://observe.live/mcp
Claude (web, desktop and mobile)
Settings → Connectors → Add custom connector. Name it Observe, paste the server URL and click Add, then Connect.
Claude Code
claude mcp add --transport http observe https://observe.live/mcp
Then run /mcp in Claude Code and choose Authenticate.
ChatGPT
Settings → Apps & Connectors → Advanced settings: turn on developer mode, then Create. Paste the server URL, choose OAuth and create it; ChatGPT opens the sign-in.
Cursor
Add this to ~/.cursor/mcp.json (or .cursor/mcp.json in a project); Cursor then asks you to log in to the server.
{
"mcpServers": {
"observe": { "url": "https://observe.live/mcp" }
}
}
VS Code
Add this to .vscode/mcp.json (or run MCP: Add Server → HTTP); VS Code asks you to sign in on first use.
{
"servers": {
"observe": { "type": "http", "url": "https://observe.live/mcp" }
}
}
- Scopes: mcp:use and monitors:read give read access; monitors:write lets the app make changes if you choose read and write. Your role still applies, as in the dashboard.
- Access tokens last an hour and refresh on their own; disconnecting the app or leaving the organization stops it at once.
- Discovery for client authors:
https://observe.live/.well-known/oauth-protected-resource/mcpandhttps://observe.live/.well-known/oauth-authorization-server(OAuth 2.1, PKCE S256, dynamic client registration).
Connect with an API token
For clients without OAuth, scripts and CI.
- Create a token under Profile → API & data. Read-only is enough to ask questions; choose read and write if the assistant should change things. You can limit it to specific monitors.
- Add the server to your client with the token as a bearer header. Server URL:
https://observe.live/mcp
Claude Code
claude mcp add --transport http observe https://observe.live/mcp \
--header "Authorization: Bearer YOUR_TOKEN"
Cursor, VS Code and other HTTP clients
Add this to the client's MCP config (Cursor: ~/.cursor/mcp.json or .cursor/mcp.json in a project).
{
"mcpServers": {
"observe": {
"type": "http",
"url": "https://observe.live/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
Claude Desktop
Claude Desktop starts local servers from its config (Settings → Developer → Edit config), so bridge to the HTTP server with mcp-remote (needs Node.js):
{
"mcpServers": {
"observe": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://observe.live/mcp", "--header", "Authorization: Bearer YOUR_TOKEN"]
}
}
}
- Same permissions as the API and the dashboard: viewers cannot change monitors, and maintenance, status pages, notification channels and members need an admin.
- A token limited to specific monitors only sees those monitors and cannot use organization tools.
- 120 tool calls a minute per token. Assistants are told to confirm before deleting a monitor or posting a public status update, and tools that change or delete data say so in their annotations.
Tools
31 tools. Read tools work with any token; write tools need read and write.
list-organizations
read
List the organizations you belong to, with your role, plan and monitor limit.
Try: “Which organizations am I in, and what plan is each on?”
update-org-settings
write
destructive
Change an organization's settings: name, monthly report, business hours, the main status page and the health check. Only the fields you pass change; the result is every setting. Admins only; needs write access.
Arguments:
organization_id*,
settings*
Try: “Turn on our status page at status slug acme and set business hours to 08:00 to 18:00.”
list-monitors
read
List monitors (websites) you can see, with their current up/down, incident, maintenance and snooze status.
Arguments:
organization_id
Try: “Which of my monitors are down right now?”
get-monitor
read
Get one monitor: its settings, current status, 30-day uptime and the five most recent incidents.
Arguments:
monitor_id*
Try: “How has shop.example.com done over the last 30 days?”
create-monitor
write
Create a website monitor. Uptime and SSL checks start straight away; pass settings to turn on other check types (crawl, Lighthouse, domain, DNS, sitemap, robots, heartbeat …). Needs write access.
Arguments:
url*,
organization_id,
label,
tags,
critical,
settings
Try: “Monitor https://docs.example.com every minute and check its domain expiry.”
update-monitor
write
destructive
Change a monitor: its URL, label, tags, or any check type's settings. Only the fields you pass change; turning a check type on queues its first check. Needs write access.
Arguments:
monitor_id*,
url,
label,
tags,
critical,
settings
Try: “Make the checkout monitor look for the text "Add to cart" and alert after 3 failures.”
enable-check
write
destructive
Turn one check type on or off for a monitor, optionally with its settings (e.g. crawl_max_pages, lighthouse_strategy). Turning it on queues the first check. Needs write access.
Arguments:
monitor_id*,
type*,
enabled,
settings
Try: “Turn on the weekly crawl for the blog, at most 500 pages.”
delete-monitor
write
destructive
Delete a monitor and its history. This cannot be undone; confirm with the user first. Needs write access.
Arguments:
monitor_id*
Try: “Delete the staging monitor.”
run-check
write
Queue a check now: uptime by default, or ssl, crawl, lighthouse, sitemap, robots, domain, dns-record, dns-blocklist or reverse-monitor. The type must be enabled. Results show in get-check-results within a minute or so (a crawl takes longer). Needs write access.
Arguments:
monitor_id*,
type
Try: “Run a Lighthouse check on the homepage now.”
get-check-results
read
Get recent check results for a monitor, newest first: ssl, domain, dns (record snapshots), lighthouse, crawls (find crawl IDs for get-crawl-pages), heartbeats, sitemap, robots, blocklist or reverse. Uptime checks are in get-monitor and get-sla-report.
Arguments:
monitor_id*,
type*,
limit
Try: “When does the SSL certificate for api.example.com expire?”
get-crawl-pages
read
List the pages a crawl visited, e.g. only broken links (filter=broken) with the page each was found on. Find crawl IDs with get-check-results type=crawls.
Arguments:
monitor_id*,
crawl_id*,
filter,
search,
limit
Try: “List the broken links from the last crawl and the pages they are on.”
get-sla-report
read
Uptime report for a monitor over 24h, 7d (default), 30d or a custom range: uptime %, downtime, MTTR, response-time percentiles and each downtime period.
Arguments:
monitor_id*,
range,
start,
end
Try: “What was our uptime and MTTR last month?”
list-incidents
read
List a monitor's incidents (downtime and heartbeat failures), newest first, with private team notes.
Arguments:
monitor_id*
Try: “Show the incidents on the API monitor this week.”
list-organization-incidents
read
Incidents across every monitor you can see in an organization, open ones first. Filter by status (open or resolved).
Arguments:
organization_id*,
status
Try: “What is open right now across the organization?”
acknowledge-incident
write
destructive
Acknowledge an open incident. Further alerts for it are muted until it resolves.
Arguments:
monitor_id*,
incident_id*
Try: “Acknowledge the checkout outage; I am on it.”
add-incident-note
write
Add a private note to an incident's timeline (findings, actions taken). Only the team sees notes; they are never sent in alerts.
Arguments:
monitor_id*,
incident_id*,
body*
Try: “Add a note to that incident: "Rolled back deploy 4812".”
snooze-alerts
write
destructive
Snooze a monitor's alerts for 1, 4, 24 or 168 hours, or pass 0 to resume alerts now.
Arguments:
monitor_id*,
hours*
Try: “Snooze the marketing site for 4 hours.”
generate-report
read
Run a report across an organization's monitors: uptime %, downtime, incidents, MTTR, response times (avg, p50, p95), SSL and domain expiry, Lighthouse, crawl health, missed heartbeats and DNS record changes (nameserver and email auth changes flagged), per monitor and per group/tag/team, optionally filtered by tag or team and against the previous period. Pass report_id to run a saved report, or a scope, range and sections.
Arguments:
organization_id,
report_id,
scope,
scope_ids,
range,
custom_start,
custom_end,
compare,
group_by,
filter_tags,
filter_team_ids,
sections
Try: “Report uptime and p95 response time for the client-a group over the last 30 days, compared with the month before.”
list-maintenance-windows
read
Maintenance windows that ended in the last 30 days or are still to come. Alerts are muted for the monitors they cover while they run.
Arguments:
organization_id*
Try: “Is any maintenance scheduled this week?”
schedule-maintenance
write
Schedule a maintenance window: alerts are muted and status pages show maintenance for the monitors it covers (all of them unless you pass monitor_ids, team_id or tags). Admins only; needs write access.
Arguments:
organization_id*,
title*,
starts_at*,
ends_at*,
description,
monitor_ids,
team_id,
tags,
recurrence,
timezone
Try: “Schedule maintenance for the database monitors tonight 22:00 to 23:00 UTC.”
cancel-maintenance
write
destructive
Cancel a maintenance window. Admins only; needs write access.
Arguments:
organization_id*,
window_id*
Try: “Cancel tonight's maintenance window.”
list-status-pages
read
An organization's additional status pages (title, slug, URL, monitors) and its latest status page updates.
Arguments:
organization_id*
Try: “What did we last post on the status page?”
post-status-update
write
Post an update to the organization's status page (or one additional page with status_page_id), e.g. an incident write-up or scheduled work. It is public as soon as it is posted; confirm the wording with the user. Admins only; needs write access.
Arguments:
organization_id*,
title*,
body*,
severity,
pinned,
status_page_id
Try: “Post a status update: we are investigating slow checkout.”
list-notification-channels
read
Where an organization's alerts go: email, Slack, Discord, Teams, webhooks, PagerDuty, Opsgenie, Jira Service Management … with the events each receives. Secrets are not shown. Admins only.
Arguments:
organization_id*
Try: “Where do our alerts go?”
send-test-notification
write
Send a sample alert through one notification channel to check it is set up right. 5 a minute per channel. Admins only; needs write access.
Arguments:
organization_id*,
channel_id*
Try: “Send a test alert to the Slack channel.”
list-servers
read
Server monitors in an organization with their heartbeat, ping, SSL and health-check state.
Arguments:
organization_id*
Try: “Have all servers sent their heartbeat?”
create-server-monitor
write
Add a server monitor: expected open or closed TCP ports, heartbeat, ping, SSL and a health URL. The checks it turns on run once now. Members (not viewers); needs write access.
Arguments:
organization_id*,
name*,
host*,
settings
Try: “Monitor db-1.example.com: port 5432 open, port 22 closed, and ping it.”
update-server-monitor
write
destructive
Change a server monitor. Only the fields you pass change; the checks it has on run once now. Needs write access and edit rights on the server.
Arguments:
organization_id*,
server_id*,
name,
host,
settings
Try: “Turn on the SSL check for web-1 on port 8443.”
delete-server-monitor
write
destructive
Delete a server monitor and its history. This cannot be undone; confirm with the user first. Needs write access.
Arguments:
organization_id*,
server_id*
Try: “Delete the old staging server monitor.”
list-teams
read
Teams in an organization and who is on each. Teams decide who sees which monitors and where team alerts go.
Arguments:
organization_id*
Try: “Who is on the platform team?”
list-members
read
People in an organization with their roles, and pending invitations. Admins only.
Arguments:
organization_id*
Try: “Who are the admins, and are any invitations pending?”
Check settings in create-monitor, update-monitor and enable-check use the API's field names; see check types and the API reference.