/**
* AI SDK Agent Example - Local (stdio)
*
* Demonstrates connecting to a local dbt-mcp server via stdio transport.
* Useful for development and testing without dbt Cloud.
*/
import { experimental_createMCPClient as createMCPClient, streamText } from "ai";
import { openai } from "@ai-sdk/openai";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
import * as readline from "readline";
import * as path from "path";
import { fileURLToPath } from "url";
const __dirname = path.dirname(fileURLToPath(import.meta.url));
const dbtMcpDir = path.resolve(__dirname, "../../..");
const envFilePath = process.env.DBT_ENV_FILE ?? path.join(dbtMcpDir, ".env");
async function main() {
// Spawn local dbt-mcp server as a subprocess
const transport = new StdioClientTransport({
command: "uvx",
args: ["--env-file", envFilePath, "dbt-mcp"],
});
// Initialize MCP client and fetch available tools
const mcpClient = await createMCPClient({ transport });
const tools = await mcpClient.tools();
console.log("Connected to local dbt MCP server");
console.log(`Available tools: ${Object.keys(tools).length}`);
console.log("Type 'quit' to exit\n");
const rl = readline.createInterface({
input: process.stdin,
output: process.stdout,
});
const prompt = (query: string): Promise<string> =>
new Promise((resolve) => rl.question(query, resolve));
try {
// Main chat loop
while (true) {
const userInput = await prompt("You: ");
if (userInput.toLowerCase() === "quit") {
break;
}
if (!userInput.trim()) {
continue;
}
process.stdout.write("Assistant: ");
// Stream response from LLM, allowing it to call dbt-mcp tools
const result = streamText({
model: openai("gpt-4o"),
tools,
maxSteps: 10,
system: "You are a helpful assistant with access to dbt tools. Use them to answer questions about the user's dbt project, metrics, models, and data.",
prompt: userInput,
onStepFinish: ({ toolCalls, toolResults }) => {
if (toolCalls?.length) {
for (const call of toolCalls) {
console.log(`\n[Tool: ${call.toolName}]`);
}
}
if (toolResults?.length) {
for (const result of toolResults) {
const preview = JSON.stringify(result.result).slice(0, 200);
console.log(`[Result: ${preview}${preview.length >= 200 ? "..." : ""}]`);
}
process.stdout.write("\nAssistant: ");
}
},
});
try {
for await (const chunk of (await result).textStream) {
process.stdout.write(chunk);
}
} catch (err) {
console.error(`\n[Error: ${err instanceof Error ? err.message : err}]`);
}
console.log("\n");
}
} finally {
rl.close();
await mcpClient.close();
console.log("Goodbye!");
}
}
main().catch(console.error);
