vercel/ai · error · MCPClientError

Unsupported request schema. Only ElicitationRequestSchema is

Error message

Unsupported request schema. Only ElicitationRequestSchema is supported.

What it means

onArrowTools/onToolRegistration (mcp-client.ts) accepts an elicitation handler, but only the MCP spec's ElicitationRequestSchema is supported as the request schema. Passing any other schema value (e.g. a different SDK schema object or a mistyped import) is rejected immediately with MCPClientError so unsupported request kinds are never silently registered.

Source

Thrown at packages/mcp/src/tool/mcp-client.ts:1396

    return this.getPromptInternal({ name, args, options });
  }

  complete(
    args: CompleteRequestParams & {
      options?: RequestOptions;
    },
  ): Promise<CompleteResult> {
    return this.completeInternal(args);
  }

  onElicitationRequest(
    schema: typeof ElicitationRequestSchema,
    handler: (
      request: ElicitationRequest,
    ) => Promise<ElicitResult> | ElicitResult,
  ): void {
    if (schema !== ElicitationRequestSchema) {
      throw new MCPClientError({
        message:
          'Unsupported request schema. Only ElicitationRequestSchema is supported.',
      });
    }

    this.elicitationRequestHandler = handler;
  }

  private async onRequestMessage(request: JSONRPCRequest): Promise<void> {
    try {
      if (request.method === 'ping') {
        await this.transport.send({
          jsonrpc: '2.0',
          id: request.id,
          result: {},
        });
        return;
      }

View on GitHub (pinned to 69428b1f8b)

Solutions

Import ElicitationRequestSchema from the same module version the client uses (the package's exported symbol) and pass that exact object as the first argument
Check your @ai-sdk/mcp (or ai) package version — older versions did not support elicitation; upgrade so the schema and client come from the same release
If you need a different request kind, do not reuse this API; handle other server requests at the transport level or file/await library support

Example fix

// before
import { ElicitationRequestSchema } from '@modelcontextprotocol/sdk/types.js';
client.experimental_onElicitationRequest(ElicitationRequestSchema, handler);
// after
import { ElicitationRequestSchema } from 'ai/mcp-stdio'; // same package as the client
client.experimental_onElicitationRequest(ElicitationRequestSchema, handler);

Defensive patterns

Strategy: validation

Validate before calling

import { ElicitationRequestSchema } from 'ai/mcp-stdio'; // same module the client uses
if (schema !== ElicitationRequestSchema) {
  throw new Error('elicitation handler must be registered with ElicitationRequestSchema from this package');
}

Try / catch

try {
  client.experimental_onElicitationRequest(ElicitationRequestSchema, handler);
} catch (error) {
  if (MCPClientError.isInstance(error) && error.message.includes('Unsupported request schema')) {
    // fix the schema import/version
  }
}

Prevention

Always import ElicitationRequestSchema from the same package/version as the MCP client
Do not wrap, re-export, or recreate the schema object
Pin @ai-sdk versions so client and schema come from the same release

When it happens

Trigger: Calling client.experimental_onElicitationRequest (the elicitation handler registration API on DefaultMCPClient) with a first argument that is not the exact ElicitationRequestSchema identity — e.g. passing a re-declared/zod schema, a schema from a different MCP SDK version, or forgetting the argument entirely.

Common situations: Copy-pasting handler registration code from SDKs (like the official TypeScript MCP SDK) that use different schema objects; importing ElicitationRequestSchema from '@modelcontextprotocol/sdk' instead of the AI SDK's mcp package so the identity check (schema !== ElicitationRequestSchema) fails; wrapping or re-exporting the schema.

Related errors

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/a077a155f9057279. Report an issue: GitHub.