Vercel AI SDK Adapter

The Vercel AI SDK adapter converts WebMCP tool definitions into Vercel AI SDK tool() format, enabling seamless integration with streamText(), generateText(), and other Vercel AI SDK functions.

Installation

bun add @thestudioxi/webmcp @thestudioxi/webmcp-adapter-vercel-ai ai zod

Usage

Basic — Convert Tools

import { webmcpToVercelAITools } from '@thestudioxi/webmcp-adapter-vercel-ai';
import { createBackendAgentClient } from '@thestudioxi/webmcp';
import { streamText } from 'ai';
import { openai } from '@ai-sdk/openai';

// Connect to WebMCP server
const agentClient = createBackendAgentClient({ transport: myTransport });
await agentClient.connect();

// Get raw WebMCP tool definitions
const rawTools = await agentClient.getAvailableTools();

// Convert to Vercel AI SDK format
const tools = webmcpToVercelAITools(
  rawTools,
  (name, args) => agentClient.executeTool(name, args)
);

// Use with streamText
const result = streamText({
  model: openai('gpt-4o'),
  messages: [{ role: 'user', content: 'What text is on the page?' }],
  tools,
});

Shortcut — Using WebMCPAgentClient

The backend SDK provides a convenience method that does the conversion automatically:

import { createBackendAgentClient } from '@thestudioxi/webmcp';
import { streamText } from 'ai';

const agent = createBackendAgentClient({ transport: myTransport });
await agent.connect();

// One-liner: get tools already in Vercel AI SDK format
const tools = await agent.getVercelAITools();

const result = streamText({
  model: myModel,
  messages,
  tools,
});

How It Works

The adapter performs two conversions:

1. JSON Schema → Zod: Converts each tool's inputSchema (JSON Schema format) into a Zod schema using jsonSchemaObjectToZod()
2. Tool Definition → Vercel AI Tool: Wraps the Zod schema and description into the format expected by Vercel AI SDK's tool() function

Supported JSON Schema Features

JSON Schema	Zod Equivalent
type: 'string'	z.string()
type: 'number' / type: 'integer'	z.number()
type: 'boolean'	z.boolean()
type: 'array'	z.array()
type: 'object'	z.object() (recursive)
enum: [...]	z.enum() or z.union()
const: value	z.literal()
description	.describe() on any type
required	Required vs .optional()

API

webmcpToVercelAITools(tools, execute?)

Parameter	Type	Description
tools	WebMCPToolDefinition[]	Array of WebMCP tool definitions
execute	(name, args) => Promise<any>	Optional executor function

Returns: Record<string, VercelAITool> — tool map compatible with Vercel AI SDK

jsonSchemaObjectToZod(inputSchema)

Parameter	Type	Description
inputSchema	WebMCPToolDefinition['inputSchema']	JSON Schema object spec

Returns: z.ZodObject<any> — equivalent Zod schema

BYOK (Bring Your Own Key)

WebMCP does not provide LLM API keys. You must supply your own API key from your preferred provider (OpenAI, Google AI, Anthropic, etc.) via environment variables.