Slack Tools

npm version License: MIT

AI SDK tools for Slack. Typed, structured tool calls for AI agents. No MCP transport required.

Installation

npm install slack-tools
# or
yarn add slack-tools
# or
pnpm add slack-tools
# or
bun add slack-tools

Quick Start

import { generateText } from "ai";
import { createSlackTools } from "slack-tools";

const tools = createSlackTools(process.env.SLACK_USER_TOKEN);

const { text } = await generateText({
  model: "anthropic/claude-sonnet-4-5",
  tools,
  prompt: "Search for messages about the Q1 roadmap in #general",
});

The model string uses the Vercel AI Gateway, which provides access to OpenAI, Anthropic, Google, and other providers through a single API.

API Reference

createSlackTools

Returns all available Slack tools, ready to pass to generateText, streamText, or any AI SDK agent.

import { createSlackTools } from "slack-tools";

const tools = createSlackTools(process.env.SLACK_USER_TOKEN);

Parameters

Options

// Require approval for all tools
const tools = createSlackTools(process.env.SLACK_USER_TOKEN, {
  needsApproval: true,
});

// Require approval only for write operations
const tools = createSlackTools(process.env.SLACK_USER_TOKEN, {
  needsApproval: [
    "slack_send_message",
    "slack_schedule_message",
    "slack_create_canvas",
    "slack_join_channel",
    "slack_delete_message",
  ],
});

Returns

An object containing the following tools:

Each tool includes inputSchema, outputSchema, execute, and toModelOutput.

Required Scopes

Your Slack app needs the following user token scopes:

chat:write, canvases:read, canvases:write, channels:history, channels:join, groups:history, im:history, mpim:history, search:read.public, search:read.private, search:read.mpim, search:read.im, search:read.files, search:read.users, users:read, users:read.email

Comparison with Slack MCP

This package was built by running each tool against the Slack MCP server and comparing responses side-by-side. The goal was to match the MCP's behavior as closely as the public Slack API allows.

The MCP returns all results as stringified text inside content[0].text. This package returns typed objects from execute and provides toModelOutput for model-friendly text formatting, making it easy to programmatically compose follow-up actions from tool results.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and questions:

• Check the AI SDK documentation
• Review the Slack Web API documentation
• Open an issue in this repository

License

MIT