Rootly MCP Server: incidents and on-call from an agent
Rootly MCP Server
Rootly's official MCP server: an agent finds and updates incidents, alerts, on-call and services; toolsets trim the context
- 45 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Without readonly=true an agent creates and changes incidents, alerts, on-call schedules and status pages
- The API token and OAuth access open incident data and on-call staff details
Install
- Works in
Manual install
claude mcp add --transport http rootly https://mcp.rootly.com/mcpThe command from the README, sign-in through OAuth2 in a browser.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Without readonly=true an agent creates and changes incidents, alerts, on-call schedules and status pages
- The API token and OAuth access open incident data and on-call staff details
This is third-party code. Review the repository files before installing.
What it does
The server connects an agent to Rootly, an incident management platform. It is hosted as a remote server at https://mcp.rootly.com/mcp with OAuth2 sign-in, and an API token is only a fallback. There are about 280 tools in all, and the full surface takes around 125 thousand tokens of context, so they are split into toolsets. The 29-tool core set handles everyday work: finding and updating incidents, alerts and who is on call. The other sets cover action items, retrospectives, alerts and routing, schedules and escalations, workflows and their steps, the service and team catalog, status pages, configuration, users, AI SRE investigations, automations, observability and the audit log. You select sets with ?toolsets=core,workflows or the X-MCP-Toolsets header. The readonly=true parameter hides and refuses all write tools. There is a Code Mode at /mcp-codemode, and a Gemini CLI extension.
Who it is for. SRE engineers and on-call teams who run incidents in Rootly and want to ask an agent about current alerts, who is on call and retrospectives.
Good fit when
- An alert fired overnight and you need to know quickly who is on call and which incidents are open
- After an incident you need a retrospective and a month's action items for a service
- You want to find similar incidents and see how they were closed before
- On-call schedules and escalation policies must be checked against the team roster
- The agent gets read-only access to working data, with writes off through readonly=true
Not a fit when
- You need all 280 tools at once: the context will overload, pick sets for the task
- Your team does not use Rootly: the server works only with its API
- The agent must not set up things Rootly will run later: leave out the workflow_tasks and automations sets
- The client still uses the SSE address: it will be removed after November 20, 2026
Example request
Show open incidents for the payments service and tell me who is on call and which action items remain from last week.MCP
- Transport
- http
- Authentication
- OAuth
Environment variables
ROOTLY_API_TOKENsecret- Rootly API token for self-hosting, not needed with OAuth2
ROOTLY_MCP_TOOLSETS- Default toolsets for a client that makes no selection
Limitations
You need a Rootly account, with sign-in through OAuth2 in a browser or an API token in the Authorization header. The full 280-tool surface overloads the model, and the README advises starting with the 29-tool core set.
The SSE address is deprecated and will be switched off after November 20, 2026, so move to Streamable HTTP. The workflow_tasks and automations sets configure actions Rootly will run later by itself: a Slack post, paging on-call, opening a ticket, calling a URL. Without readonly=true an agent can change working incident data.
How to disable. Remove the server with claude mcp remove rootly. Revoke access in Rootly settings, and kill the API token there too if you used one.
FAQ
How do I connect Rootly MCP to Claude Code?
With claude mcp add --transport http rootly https://mcp.rootly.com/mcp. The client opens a browser window for OAuth2 sign-in, no API token needed. For a token add an Authorization header with Bearer.
Why does the server have so many toolsets?
The full list of 280 tools takes around 125 thousand tokens, so a client loads sets for the task. The core set of 29 tools is recommended, and others are added with ?toolsets= or the X-MCP-Toolsets header.
How do I stop the agent from changing data?
Add readonly=true to the URL or the X-MCP-Readonly: true header. Write tools are hidden and refused, and the server rejects any non-GET request. It is also wise to leave out the workflow_tasks and automations sets.
Which toolsets exist?
Fourteen: core, incidents, alerts, oncall, workflows, workflow_tasks, catalog, status_pages, config, users, ai, automations, observability and audit. The server refuses a tool outside the selected sets and names the set that holds it.
What happens to the SSE address?
It is deprecated, like the --transport sse option of a self-hosted server, and will be removed after November 20, 2026. Replace it with https://mcp.rootly.com/mcp or the client will stop connecting.
Discussion
No questions or reviews yet.
Related
MCP serversThe built-in OneUptime MCP: an agent works incidents, reads logs and traces, acknowledges alerts and sets up monitors.
The RootCX CLI and skill: an app written by an agent gets a database, SSO, roles, audit and deployment on one server.
Self-hosted ops platform with an SRE AI agent and an MCP endpoint that lets an external AI client triage alerts and check hosts
MCP Slim Guard sits between an agent and your MCP servers: it validates calls against schemas and returns big results in parts.
GitHub's official MCP server: code, issues, pull requests, Actions and security alerts straight from the agent
An agent runs servers through a catalog of declared actions: policy decides what runs and what waits for a person
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Rootly MCP Server: incidents and on-call from an agent (this card) | yes | OAuth | 45 | |
| Sentry MCP | no | API key | 961 | errors and traces in Sentry, with no on-call management |
| Hermes Incident Commander | no | 75 | an incident commander agent without the Rootly platform |
README in short
The README starts with a quick start on the hosted server and describes the addresses, a table of toolsets and selecting a set by parameter or header. Next come connection examples for general clients, Claude Code, Gemini CLI and other agents, plus self-hosting settings with environment variables.