Agent-native incident response
WebMCP: initializing

Agent

Every capability CrisisDesk exposes to an AI agent, registered live via document.modelContext.registerTool.

WebMCP Status
InitializingThis browser does not implement WebMCP yet, so CrisisDesk is using a same-shaped local polyfill (lib/webmcp/polyfill.ts) so the demo still works end to end — never presented as native.
Registered Tools (10)
get_incident

Retrieve structured details for an incident by its ID, including status, severity, timeline, and affected services.

get_service_status

Get the current health, version, uptime, and dependencies of a service.

get_metrics

Get error rate, latency, request volume, CPU, and memory metrics for a service over a recent time window.

get_logs

Get recent structured log entries for a service, optionally filtered by severity.

get_recent_deployments

Get the recent deployment history for a service, most recent first.

get_dependencies

Get the upstream services a service depends on, and the downstream services that depend on it.

correlate_evidencestate-changing

Record one or more correlated signals as evidence on an incident, after reviewing logs, metrics, and deployments. Call this once you have identified concrete signals worth attaching to the investigation record.

create_root_cause_hypothesisstate-changing

Record the agent's root cause hypothesis for an incident, with a confidence score and a concise, decision-relevant reasoning summary. Do not include hidden chain-of-thought — only the evidence-backed summary.

create_remediation_planstate-changing

Propose a structured remediation plan for an incident. This only records a proposal — it does not change any system state.

execute_remediationstate-changing

Execute a previously proposed remediation step against the simulated environment. HIGH IMPACT: this always pauses for explicit human approval in the UI before anything runs, and it never touches real infrastructure — only the CrisisDesk simulation.

Recent Tool Activity
No tool calls yet.