Skip to content
dinobase

Vercel AI SDK

Dinobase integrates with the Vercel AI SDK via MCP. No adapter code needed — the AI SDK’s MCP client connects directly to Dinobase’s MCP server.

Install

Section titled “Install”
Terminal window
npm install ai @ai-sdk/anthropic @ai-sdk/mcp
pip install dinobase

Set up Dinobase with your data sources:

Terminal window
dinobase init
dinobase add stripe --api-key sk_test_...
dinobase sync

See Connecting Sources for the full list of 100+ supported sources, and Syncing & Scheduling for background sync options.

Next.js API Route

Section titled “Next.js API Route”
app/api/chat/route.ts
import { streamText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';
import { createMCPClient } from '@ai-sdk/mcp';
import { Experimental_StdioMCPTransport } from '@ai-sdk/mcp/mcp-stdio';


export async function POST(req: Request) {
  const { messages } = await req.json();


  const dinobase = await createMCPClient({
    transport: new Experimental_StdioMCPTransport({
      command: 'dinobase',
      args: ['serve'],
    }),
  });


  const tools = await dinobase.tools();


  const result = streamText({
    model: anthropic('claude-sonnet-4-6'),
    messages,
    tools,
  });


  result.onFinish(() => dinobase.close());


  return result.toDataStreamResponse();
}

The AI SDK automatically discovers all 7 Dinobase MCP tools.

Node.js Script

Section titled “Node.js Script”

For scripts and CLI tools, use generateText():

import { generateText } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';
import { createMCPClient } from '@ai-sdk/mcp';
import { Experimental_StdioMCPTransport } from '@ai-sdk/mcp/mcp-stdio';


const dinobase = await createMCPClient({
  transport: new Experimental_StdioMCPTransport({
    command: 'dinobase',
    args: ['serve'],
  }),
});


const tools = await dinobase.tools();


const { text } = await generateText({
  model: anthropic('claude-sonnet-4-6'),
  prompt: 'Which customers have overdue invoices?',
  tools,
  maxSteps: 5,
});


console.log(text);
await dinobase.close();

Available Tools

Section titled “Available Tools”

The MCP server exposes these tools to the AI model:

ToolDescription
queryExecute SQL queries (DuckDB dialect)
describeGet table schema, column types, and sample data
list_sourcesList all connected sources with row counts and freshness
refreshRe-sync a data source to get fresh data
confirmExecute a pending mutation (write-back to source API)
confirm_batchExecute multiple pending mutations
cancelCancel a pending mutation

Background Sync

Section titled “Background Sync”

To keep data fresh while the MCP server runs:

const dinobase = await createMCPClient({
  transport: new Experimental_StdioMCPTransport({
    command: 'dinobase',
    args: ['serve', '--sync', '--sync-interval', '30m'],
  }),
});

Notes

Section titled “Notes”
  • Experimental_StdioMCPTransport runs in Node.js only (not edge runtime or browser)
  • Always close the MCP client when done to clean up the subprocess
  • The model sees dynamic instructions computed from your actual database state — it knows what sources and tables are available

Next steps

Section titled “Next steps”