AI agent integrations for TypeScript

The canonical @agentskit/integrations package contains 50 service descriptors. This page covers the detailed guides available in the tools documentation; the for-agents package guide describes the full descriptor registry and projection contract. The canonical integration catalog keeps both surfaces discoverable without creating placeholder pages for services that do not yet have a dedicated guide.

#Categories

#Usage

Import whole integrations for the full tool set, or cherry-pick sub-tools:

import { github, slack, stripeCreatePaymentIntent } from '@agentskit/tools/integrations'
import { createRuntime } from '@agentskit/runtime'

const runtime = createRuntime({
  adapter,
  tools: [
    ...github({ token: process.env.GITHUB_TOKEN! }),
    ...slack({ token: process.env.SLACK_BOT_TOKEN! }),
    stripeCreatePaymentIntent({ apiKey: process.env.STRIPE_API_KEY! }),
  ],
})

#Credentials

Every integration reads a token or API key from its config. No process-env magic, no globals — you pass what you want, where you want.

#Conventions

• Sub-tools are named <integration><Verb> (e.g. githubCreateIssue).
• Bundled tool exported under the integration name (e.g. github) returns an array of sub-tools.
• Config types exported as <Integration>Config.
• Errors wrap provider 4xx/5xx with ToolError so runtime sees a consistent shape.
• Mandatory sandbox — wrap risky integrations (shell, postgres write, github create) via createMandatorySandbox.

#Related

• Authoring tools — wrap what's missing.
• Built-in tools — webSearch, fetchUrl, filesystem, shell.
• MCP bridge — consume or publish tool sets over MCP.
• Recipes: integrations · more integrations.