Ask questions about incident.io
A managed agent for read-only incident.io incident and on-call analysis.
- chat
- incidentio-answerer
- incidentio-readonly
incidentio-answerer-pool
- incidentio-answerer-model
incidentio
What this template gives you
Interface
A chat window, in your browser
Use the included interface to send a request to this template.
Gateway
incidentio-readonly gateway
This template sends its requests through incidentio-readonly gateway, where its access rules and limits apply.
Model
incidentio-answerer-model
This template uses incidentio-answerer-model through the incidentio-answerer-pool pool.
Usage
Hand this template to an AI agent to set up this template
Node.js and npm to run npx.
Dome CLI, installed with brew install dome-systems/tap/dome, then signed in with dome auth login.
Docker Compose to run the companion app locally.
- 1
Create a workspace
dome workspaces create --name <workspace-name>The CLI creates and selects a workspace context for you.
- 2
Confirm the workspace context
dome context listThe workspace you just created is already selected.
- 3
Pull the template
npx @domesystems/templates pull dome-systems/ai-agent-for-incidentio@0.1.0 --dir ai-agent-for-incidentioDownloads this immutable release and verifies its digest.
- 4
Run the template
npx @domesystems/templates up --dir ai-agent-for-incidentioImports it into Dome and starts the companion app.
Dome import variables
npx @domesystems/templates up imports the template into Dome and prompts for required values. For unattended runs, set them as DOME_IMPORT_VAR_… environment variables.
Required values
DOME_IMPORT_VAR_anthropic_api_key
Setup guide
Follow this template's setup guidance for agent variables and upstream connections.
# Configure incident.io
incident.io's remote MCP server uses OAuth with Dynamic Client Registration.
It is available only on incident.io Team, Pro, and Enterprise plans.
Before importing this template:
- An incident.io administrator must search **Settings** for **MCP**. If there is
no MCP setting, the remote server is not provisioned for the organization;
do not start OAuth. Ask the organization owner or incident.io Support to
confirm plan eligibility and enable the feature.
- In **Settings → MCP**, enable the remote MCP server and add
`api.domesystems.ai` to **Allowed redirect domains**.
- `anthropic_api_key`: enter the Anthropic key used by this agent's model.
- Use a dedicated incident.io test user rather than an organization owner.
- When OAuth opens, select the intended incident.io organization and choose
**Read only**. The template can then inspect the same incident, alert,
escalation, on-call, and catalog data that user can see, without making
changes.
- Dome registers its OAuth client automatically; do not create an incident.io
client ID or client secret.
After saving, complete OAuth as the dedicated user. Confirm that
`identity_show` reports `access_level: read_only`, then test a known incident
and an out-of-boundary team or private incident.
If consent says **“The MCP server isn't available for [organization]”**, the
remote MCP server is disabled or has not been provisioned. Return to
**Settings → MCP**; if that setting is absent, ask the organization owner or
incident.io Support to enable it before retrying.
For a non-interactive connection, incident.io also supports API keys. Create
one under **Settings → API keys**, grant only the required account- or
team-scoped read permissions, store the value immediately (it is shown once),
and use it only with an MCP client configured for API-key authentication.Runtime environment
These variables are required to run the agent. During pull, .env.example is copied to .env when available. If they’re missing, npx @domesystems/templates up retrieves them from Dome and saves them in .env. Pass --provision to refresh them.
No .env.example is included in this release.Agent instructions
Answer questions using the connected incident.io organization. Search and summarize
incidents, alerts, escalations, schedules, catalog data, and related operational
context without changing anything. Never create, update, acknowledge, resolve,
merge, or delete incidents; modify on-call schedules or escalation paths; trigger
workflows; start or steer investigations; or publish status-page updates. If asked
to make a change, explain that this is a read-only template and offer a reviewable
plan for a person to apply.