Vercel AI SDK — Nitrosend

@nitrosend/ai-sdk exposes Nitrosend's MCP tool surface — sending email, managing contacts, building flows, running campaigns — to applications built on the Vercel AI SDK. It is a typed wrapper over the Nitrosend remote MCP server and is not a model package.

Install

npm install @nitrosend/ai-sdk ai @ai-sdk/mcp zod

The examples below also use @ai-sdk/openai for the model. Substitute your preferred AI SDK model package (@ai-sdk/anthropic, @ai-sdk/google, etc.):

npm install @ai-sdk/openai

Authenticate

Get a key from Settings → API Keys and export it. The package reads NITROSEND_API_KEY automatically, or you can pass apiKey explicitly.

export NITROSEND_API_KEY="nskey_live_<your-api-key>"

Quickstart — generateText

stopWhen: isStepCount(N) is required for tool-calling workflows: AI SDK defaults to a single step, so without it the call stops as soon as the model invokes a Nitrosend tool and you never see the final summary text.

import { generateText, isStepCount } from 'ai';
import { openai } from '@ai-sdk/openai';
import { withNitrosendTools } from '@nitrosend/ai-sdk';

const result = await withNitrosendTools({}, async ({ tools }) => {
  return generateText({
    model: openai('gpt-4o'),
    tools,
    stopWhen: isStepCount(5),
    prompt: 'Send a welcome email to founder@acme.com.',
  });
});

Streaming — streamText

import { streamText, isStepCount } from 'ai';
import { openai } from '@ai-sdk/openai';
import { withNitrosendTools } from '@nitrosend/ai-sdk';

await withNitrosendTools({}, async ({ tools }) => {
  const stream = streamText({
    model: openai('gpt-4o'),
    tools,
    stopWhen: isStepCount(5),
    prompt: 'Add founder@acme.com to the "Power users" list and confirm.',
  });

  for await (const part of stream.textStream) {
    process.stdout.write(part);
  }

  await stream.consumeStream();
});

withNitrosendTools opens the MCP transport, hands you the toolset, and closes the transport in a finally block when your callback resolves or throws — including when streamText is awaited inside.

Lifecycle and subsets

For longer-lived processes, hold the client yourself:

import { nitrosend } from '@nitrosend/ai-sdk';

const { tools, client, close } = await nitrosend();
try {
  // …pass `tools` to generateText/streamText, or use `client` directly…
} finally {
  await close();
}

Filter the toolset to only what your agent should reach for — types narrow to match. Use withNitrosendTools so the transport closes after use:

import { withNitrosendTools } from '@nitrosend/ai-sdk';

await withNitrosendTools(
  { tools: ['nitro_get_status', 'nitro_send_message'] },
  async ({ tools }) => {
    // tools is typed to exactly { nitro_get_status, nitro_send_message }
  },
);

Errors

The wrapper does not retry. All failures throw NitrosendAISDKError with a stable .code and the original error preserved on .cause:

AUTH_MISSING — no credential was found in options or env.
AUTH_INVALID_PREFIX — apiKey did not start with nskey_live_ / nskey_test_.
CLIENT_INIT_FAILED — MCP transport could not connect (check URL, network, key).
TOOLS_LIST_FAILED — server rejected the tools list call.

REST API — Nitrosend HTTP API used by @nitrosend/sdk.
MCP overview — connecting Claude Desktop / Cursor / IDEs via the @nitrosend/mcp stdio bridge.
Tools Registry submission — public AI SDK tool directory; this package targets that listing.

The companion native Vercel listing (one-click install, secret rotation, billing) is tracked in a separate Nitrosend spec and is not yet shipped.