Slack MCP Client

What is Slack MCP Client?

A Slack bot with an MCP client built in TypeScript, enabling users to execute MCP server tools directly from Slack conversations. It runs locally and requires Node 22, pnpm 10, an OpenAI API key, and at least one MCP server accessible via stdio, SSE, or streamable HTTP.

How to use Slack MCP Client?

Install dependencies with pnpm install, copy configuration files from mcp-example.json to mcp.json and .env.example to .env, then populate with Slack tokens (signing secret, bot token, app-level token) and the auth redirect URL. Optionally expose localhost via ngrok or enable socket mode. Run pnpm run dev and open a new thread with the bot to start interacting.

Key features of Slack MCP Client

• Supports HTTP Streamable MCP servers (protocol 2025-03-26)
• Supports Stdio MCP servers (protocol 2025-03-26)
• OAuth‑based authorization flow (protocol 2025-03-26)
• SSE MCP servers (previous protocol version)
• MCP Tools support (more capabilities coming soon)
• Runs as a Slack AI app using the Bolt framework

Use cases of Slack MCP Client

• Execute MCP server tools (e.g., LinkedIn actions) directly from Slack threads
• Automate workflows by connecting internal MCP servers to Slack conversations
• Prototype and test MCP servers with an interactive Slack interface

FAQ from Slack MCP Client

What does Slack MCP Client do?

It acts as a client for MCP servers, allowing users to invoke tools exposed by those servers via a Slack bot interface.

Which MCP transport protocols are supported?

HTTP streamable and stdio as defined in protocol version 2025-03-26, plus SSE from the previous protocol version.

Do I need an OpenAI API key?

Yes, the bot requires an OpenAI API key to function.

Is Slack MCP Client free?

The project is MIT licensed, so the software is free to use. You do need your own OpenAI API key and a Slack workspace.

What are the system requirements?

Node 22 (LTS/Jod) and pnpm 10. You also need a Slack app configured with appropriate tokens and scopes.

Are there any known limitations?

Currently only MCP Tools are supported; support for resources and prompts is planned but not yet implemented.