Tools

Your agent sees three tools whatever you connect. Full schemas load only for the action the agent picks.

Tool	Purpose
`discover_actions`	Find actions across connected providers
`inspect_action`	Read the description and input schema of one action
`run_action`	Validate arguments and run one action

discover_actions

Returns ranked actions for a natural language query. Each result carries a short-lived action_ref.

Request

{ "query": "list projects" }

Response
{
  "actions": [
    {
      "action_ref": "yc_ar_v1.<ref>",
      "source_id": "neon",
      "name": "list_projects",
      "summary": "List Neon projects you own.",
      "classification": "read",
      "available": true
    }
  ]
}

inspect_action

Returns the current description and input schema for one action_ref.

Request

{ "action_ref": "yc_ar_v1.<ref>" }

run_action

Validates arguments against the provider schema and calls only that provider. Write and destructive actions follow the rules in Access and approvals.

Request

{ "action_ref": "yc_ar_v1.<ref>", "arguments": { "limit": 10 } }

Response
{
  "content": [
    { "type": "text", "text": "[{\"id\": \"<project-id>\", \"name\": \"my-project\"}]" }
  ]
}

Get started

Companion

MCP

Dashboard

CLI reference

Changelog

discover_actions

inspect_action

run_action

​discover_actions

​inspect_action

​run_action

discover_actions

inspect_action

run_action