MCP server · agent-ready
Run your on-call from your AI assistant
Oncallz is available in the Claude connector directory. Connect from Claude, Claude Code, Cursor or any MCP client with OAuth sign-in or a personal token and ask what is on fire, who is on call, or to acknowledge an incident and leave a note — with exactly the permissions you already have.
Hosted, nothing to install
One remote endpoint over Streamable HTTP. No local process, no package, no self-hosting: add the URL and sign in, or use an API token.
Your permissions, not more
Every tool calls the same REST API the mobile app uses, as you. A teammate’s private team stays invisible, and a responder cannot do what a responder could not do before.
Built for 3 a.m.
Check status, read an incident’s timeline and AI analysis, acknowledge it, hand over with a comment — all from the conversation you already have open.
Connect in two minutes
You need an Oncallz account. In Claude, open the Oncallz directory listing, connect, and approve access in Oncallz. You can also add https://oncallz.com/mcp directly to a compatible client. The following manual setup uses a personal API token:
Connect Oncallz from the Claude directory
- 1 Open Profile → API tokens and create a token with Full access. Pick an expiry — 90 days is a sensible default. The token is shown once; copy it.
- 2 Add the server to your client (below), pasting the token where it says YOUR_TOKEN.
- 3 Ask something. “What is the status?” is a good first question.
Claude Code
claude mcp add --transport http oncallz https://oncallz.com/mcp \
--header "Authorization: Bearer YOUR_TOKEN"
Cursor and other clients with an mcp.json
{
"mcpServers": {
"oncallz": {
"url": "https://oncallz.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
}
}
}
}
Check it from a terminal
curl -s https://oncallz.com/mcp \
-H "Authorization: Bearer YOUR_TOKEN" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
You should get back the list of tools. A 401 means the token is wrong, expired or revoked; a 403 means it is a read-only feed token, which cannot use MCP.
OAuth uses your Oncallz login and asks you to approve access. You can revoke the connection from your profile. The token examples above remain supported, including for the mobile REST API.
What your assistant can do
27 tools, each one a thin wrapper over an endpoint of the public API.
Status and incidents
get_status- Overall health, open incidents by severity, the newest ones and who is on call.
list_incidents- Filter by status, team or ownership; sort by urgency.
get_incident- Timeline, raw alert payload, AI analysis and related incidents.
acknowledge_incident- Take it, with an optional note on the timeline.
resolve_incident- Close it, with an optional note.
reopen_incident- Bring a resolved incident back.
comment_on_incident- Leave findings or a hand-off note for the team.
On-call
whos_on_call- Who is on call right now, overrides applied.
get_on_call_calendar- Shifts over a date range, up to 92 days.
list_teams- The teams you belong to.
whoami- Your name, role, timezone and what you may manage.
Post-mortems and status page
list_post_mortems- Find draft and published reviews across your teams.
list_owed_post_mortems- Find resolved incidents that still need a published review.
get_post_mortem_form- Read editable fields, incident facts and eligible action-item assignees.
save_post_mortem- Create or edit a review and its action items.
publish_post_mortem- Publish a review internally.
publish_post_mortem_publicly- Publish the separate customer statement on a public page.
unpublish_post_mortem_publicly- Take the public customer report down.
get_post_mortem- The write-up for an incident, with its action items.
list_post_mortem_follow_ups- Post-incident work still owed across your teams.
list_my_action_items- Action items assigned to you, soonest due first.
toggle_action_item- Tick an action item off, or back on.
get_status_page- Status-page console: component status, uptime and statements, including internal ones.
get_status_statement- One status-page incident with its update history and visibility.
create_status_incident- Record an incident, including past outages with actual start and end times.
post_status_update- Post an update, resolve or reopen a status-page incident.
edit_status_incident- Correct wording, outage times or visibility without sending notifications.
Things to ask
- “Draft a post-mortem for the checkout incident using its timeline, add follow-up action items, and keep it unpublished.”
- “Is anything on fire right now?”
- “Who is on call tonight, and who is next weekend?”
- “Summarise the open critical incidents and what the AI analysis thinks caused them.”
- “Acknowledge the checkout incident and note that I am rolling back the deploy.”
- “Which post-mortem action items are overdue, and which are mine?”
- “What does the customer status page say right now?”
- “Yesterday the API was down from 10:15 to 10:45 Istanbul time. Record a resolved status-page incident, keep details internal but downtime public, and do not notify subscribers.”
Safe by design
- Same permissions as the API: team membership and roles decide what a tool can see or change. Another team’s incident answers “not allowed”, not its title.
- Actions are labelled “MCP assistant” on the incident timeline, so nobody wonders who acknowledged what.
- Status-page tools can publish incidents and updates with your permissions. Outage status and incident details have independent visibility controls; subscriber notifications default to off. Escalating incidents, changing schedules and messaging settings remain in the app.
- OAuth connections and API tokens can be revoked from your profile. API tokens can have an expiry. MCP calls are rate limited.
- Tools are marked read-only or not, so clients can ask for confirmation before the ones that change something.
Agent-ready, not just MCP-ready
Agents should not have to guess where things are. Oncallz publishes its machine-readable surface in the places agents look:
-
MCP server card
/.well-known/mcp/server-card.jsonServer name, version, transport endpoint and authentication, for clients that discover servers.
-
API catalog (RFC 9727)
/.well-known/api-catalogA linkset pointing to the OpenAPI spec, the documentation and the health endpoint.
-
ARD capability manifest
/.well-known/ai-catalog.jsonThe MCP server and the REST API, each with example questions, for agent registries.
-
OpenAPI spec
/docs/api.jsonThe machine-readable description of the REST API the tools wrap.
-
llms.txt
/llms.txtA Markdown map of the public site for language models.
-
Markdown on request
/Send Accept: text/markdown to any public page and get Markdown back; browsers still get HTML.
-
Link headers (RFC 8288)
/Every public page advertises the catalog, spec and docs in its response headers.
-
WebMCP
/Pages register list_site_pages and open_site_page for agents running in the browser.
Frequently asked questions
What is the Oncallz MCP server?
A hosted Model Context Protocol server at oncallz.com/mcp. It lets an AI assistant such as Claude or Cursor read and act on your Oncallz data through 27 tools: check status, work incidents, see who is on call and follow up post-mortems. It acts as you, with your permissions.
Which clients work with it?
Remote Streamable HTTP clients, including Claude and Cursor, can connect with OAuth. Clients that support an Authorization header can continue using a personal API token.
Can the assistant do damage?
It can do what you can do with acknowledge, resolve, reopen and comment on incidents, and tick action items — nothing more. It can also write post-mortems, manage action items, publish internal reviews and publish or withdraw separate customer reports with your permissions.
It can create and update status-page incidents, including historical outages, when you may publish. Notifications default to off and visibility is configurable. It cannot escalate or change schedules or settings. Use a token with an expiry, and revoke it from your profile whenever you like.
Does it cost extra?
No. MCP access is included in every plan, at the same price per user as the rest of Oncallz.
Is my incident data sent to an AI provider?
Oncallz returns data to your MCP client; what happens next depends on the assistant you connected. Oncallz does not send anything to a model on its own through this server.
Can I build on the API without MCP?
Yes. The same token works as a bearer token against the REST API, which is documented with an OpenAPI spec. MCP is a convenience layer on top of it.
Related
- AI investigation Before it pages, Oncallz summarises the alert and suggests a probable cause and next step; optionally it skips the phone call for clear noise. Your own AI key.
- Post-mortems Write blameless post-mortems with timelines and action items, and publish a customer-facing incident report with one click.
- On-call scheduling Build daily, weekly or custom on-call rotations with automatic handoffs and overrides, so every alert reaches whoever is on call right now.
Give your assistant an on-call rotation
Start free and connect with OAuth or an API token.