Skip to main content
Tools are the actions agents can take. Where skills describe what to do, tools are how it gets done.

​
What a tool is

A tool is a single integration call: post a Slack message, update a DNS record, open a Jira ticket, page on-call. Tools are configured inside skills, so the same skill can be reused across agents and the same tool can appear in multiple skills.

​
Built-in tool types

Clarion ships with native tools across these categories:
  • Notifications: Slack, Teams, Discord, SMS, email, PagerDuty, phone call.
  • Ticketing: Jira, GitHub Issues.
  • DNS & infrastructure: Route 53, Cloudflare DNS, generic DNS providers.
  • Cloud: AWS (CloudTrail lookups, IAM queries), Vercel.
  • Endpoint: Huntress, CrowdStrike.
  • On-chain response: Pause actions for smart contracts.
  • Alerting: Create or update Clarion alerts from a skill.
  • Approval & clarification: Request human input before continuing.

​
Custom MCP tools

Bring your own tool over MCP. Any Model Context Protocol server you point Clarion at becomes available as a tool inside skills. Use this to integrate internal systems without writing a custom connector.

​
Approval and execution modes

You can configure how tools run from your Workspace > Tools settings. Two execution modes are available, and both can be set at the tool level or per individual action:
  • Automatic: The agent executes the tool without asking.
  • Requires approval: The agent pauses and waits for a human to approve before executing.
Some tools also expose an execution level that controls the blast radius of an action:
  • Local: Narrow scope, typically limited to a single resource or account.
  • Protected: Wider scope, gated more carefully and usually combined with approval.
  • No restriction: Full access. Combine with approval gates for sensitive actions.

​
What an execution looks like

When a tool runs during an agent investigation, Clarion records:
  • Input parameters as resolved at the time of the call.
  • Output returned by the integration.
  • External ID linking back to the system of record (Jira issue key, Slack message link, and so on).
  • Status: succeeded, failed, pending approval, or pending clarification.
  • Errors if the call failed.
Sensitive tools like DNS changes, smart contract pauses, and IAM revokes are gated behind approval requests by default. The agent pauses, the right human is notified through their preferred channel, and execution resumes after they approve.
Learn about Notifications →