Add type tests for McpToolSet and ToolSet compatibility

[Copilot]

Background

Issue #6739 reported a type mismatch between McpToolSet (from @ai-sdk/mcp) and ToolSet (from ai), preventing users from passing MCP tools directly to streamText/generateText without casting.

After investigation, the issue no longer reproduces in the current codebase—McpToolSet<'automatic'> is correctly assignable to ToolSet because Tool<unknown, CallToolResult> is assignable to Tool<any, any> in the union.

Summary

Added type tests to prevent regression:

• Verifies McpToolSet<'automatic'> is assignable to ToolSet
• Verifies individual McpToolBase<unknown> is compatible with ToolSet values
• Verifies Tool<unknown, CallToolResult> is assignable to Tool<any, any>

The test file duplicates the ToolSet type definition to avoid circular dependencies between @ai-sdk/mcp and ai packages.

Manual Verification

// Verified these compile without error:
const mcpClient = await createMCPClient({ transport: { type: 'sse', url: '...' } });
const tools = await mcpClient.tools();
const toolset: ToolSet = tools;  // ✓ No cast needed
streamText({ model, tools, prompt: 'Hello' });  // ✓ No cast needed

Checklist

• Tests have been added / updated (for bug fixes / features)
• Documentation has been added / updated (for bug fixes / features)
• A patch changeset for relevant packages has been added (for bug fixes / features - run pnpm changeset in the project root)
• I have reviewed this pull request (self-review)

Related Issues

Closes #6739

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Type mismatch: McpToolSet not assignable to ToolSet (FlexibleSchema vs FlexibleSchema)</issue_title>
<issue_description>## Description

When using experimental_createMCPClient from @ai-sdk/mcp and passing the tools to streamText, TypeScript reports a type error because McpToolSet is not assignable to ToolSet.

Reproduction

import { experimental_createMCPClient } from "@ai-sdk/mcp";
import { streamText, type ToolSet } from "ai";
import { openai } from "@ai-sdk/openai";

const mcpClient = await experimental_createMCPClient({
  transport: { type: 'http', url: 'http://localhost:3001/mcp' },
});

// ❌ Type error without cast
const tools = await mcpClient.tools();

const result = streamText({
  model: openai("gpt-4.1"),
  tools, // Error: Type 'McpToolSet' is not assignable to type 'ToolSet'
  prompt: "Hello",
});

Error Message

Type '{ [x: string]: { inputSchema: FlexibleSchema<unknown>; ... }' is not assignable to type 'ToolSet'.

Types of property 'inputSchema' are incompatible.
  Type 'FlexibleSchema<unknown>' is not assignable to type 'FlexibleSchema<never>'.
    Type '$ZodType<unknown, ...>' is not assignable to type 'StandardSchemaV1<unknown, never>'.
      Types of property 'value' are incompatible.
        Type 'unknown' is not assignable to type 'never'.

Root Cause

• mcpClient.tools() returns McpToolSet with inputSchema: FlexibleSchema<unknown>
• streamText expects ToolSet with inputSchema: FlexibleSchema<never>

The unknown vs never variance is the core issue. MCP tools have dynamic schemas (unknown at compile time), but ToolSet expects strict type inference.

Current Workaround

Cast the tools:

const tools = (await mcpClient.tools()) as ToolSet;

Expected Behavior

mcpClient.tools() should return a type compatible with streamText's tools parameter without requiring a cast.

Environment

• ai: 6.0.0-beta.124
• @ai-sdk/mcp: 0.0.11
• @ai-sdk/react: 3.0.0-beta.124
• TypeScript: 5.x</issue_description>

<agent_instructions>try to reproduce with the latest codebase and if this is still an issue, explore solutions</agent_instructions>

Comments on the Issue (you are @copilot in this section)

@aayush-kapoor could you try upgrading the mcp package to the latest beta as well?

i somehow don't observe the type error. I'm assuming it's a redline type error? </comment_new>

• Fixes Type mismatch: McpToolSet not assignable to ToolSet (FlexibleSchema<unknown> vs FlexibleSchema<never>) #10697

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.