Database

The Integrate SDK can persist OAuth provider tokens and scheduled triggers in your database instead of relying on hand-written callbacks for every CRUD operation. This enables:

• Multi-device sync: Access tokens across devices and sessions
• Server-side usage: Use tokens in API routes and background jobs
• Long-term persistence: Tokens survive browser clears and device changes
• Trigger storage: Schedule tool executions with database-backed triggers

Adapters vs callbacks

You can configure persistence in two ways:

With adapters, pass database to createMCPServer. Token and trigger CRUD are handled for you. You still provide getSessionContext to map requests to a userId (session cookie, Clerk, service headers, etc.).

import { createMCPServer, githubIntegration } from "integrate-sdk/server";
import { drizzleAdapter } from "integrate-sdk/adapters/drizzle";
import { db } from "./db";
import { providerToken, trigger } from "./db/schema";

export const { client } = createMCPServer({
  apiKey: process.env.INTEGRATE_API_KEY,
  integrations: [githubIntegration({ scopes: ["repo"] })],
  database: drizzleAdapter(db, {
    provider: "pg",
    schema: { providerToken, trigger },
  }),
  getSessionContext: async (req) => {
    const session = await auth.api.getSession({ headers: req.headers });
    return session?.user?.id ? { userId: session.user.id } : undefined;
  },
});

Explicit callbacks on createMCPServer still work and override adapter defaults when both are set.

Connected-provider tool loading

When using a database adapter, pair token storage with connected-only tool discovery:

import { getVercelAITools } from "integrate-sdk/server";

const tools = await getVercelAITools(serverClient, {
  context: { userId: session.user.id },
  connectedOnly: true,
  mode: "code",
});

Use listConnectedProviders from integrate-sdk/server when you need the connected provider list outside AI helpers.

Guides

• Drizzle: PostgreSQL, MySQL, SQLite
• Prisma: @prisma/client
• MongoDB: native driver collections
• Schema: canonical tables and indexes
• Triggers: composing onComplete and getCallbackUrl
• Hooks: cache invalidation, identity resolution, trigger authorization

Examples

Runnable Next.js examples for each database adapter live in the examples repository:

Each example connects GitHub OAuth, persists tokens with the adapter, and lists repositories from the browser.

What stays app-specific

getSessionContext cannot be fully abstracted. Each app wires its auth library or internal service headers. Optional hooks cover app-specific behavior (cache revalidation, GitHub email resolution, trigger ownership repair) without forking the adapter.

Next steps

• OAuth authorization
• Triggers (getting started)
• Built-in Integrations