MCP (Model Context Protocol)

This repository contains examples of using Pipedream's dynamic MCP server.

What makes Pipedream's MCP server unique

Unlike traditional MCP implementations that only provide a fixed set of tools, Pipedream's MCP server offers the ability to identify and load tools dynamically. This means that the tools can either be pre-defined, or discovered and configured on-demand.

This type of tool discovery is defined in the MCP spec, but most MCP clients don't support yet provide support for it.

This repo contains examples of how you, as the developer of an MCP client, can use Pipedream's app-level or dynamic MCP server.

High level

The idea is to manually reload the list of tools on each turn.

Examples

This repo includes two examples showing different approaches to integrating with Pipedream's dynamic MCP server:

• The AI SDK Example uses Vercel's AI SDK with automatic tool handling
• The OpenAI SDK Example uses OpenAI SDK directly with manual tool conversion for full control

Installation

Install dependencies:

pnpm install

Prerequisites

To set up your environment, you'll need:

1. A Pipedream account
2. A Pipedream project. Accounts connected via MCP will be stored here.
3. Pipedream OAuth credentials
4. An OpenAI API key

Set up your environment

Copy the example environment file and add your credentials:

cp .env.example .env

Learn more about environments in Pipedream Connect.

Running Examples

You can run the examples using the following commands from the root directory:

Vercel AI SDK Example

pnpm ai-sdk -u <external-user-id> "<prompt>"

OpenAI SDK Example

pnpm openai-sdk -u <external-user-id> "<prompt>"

For example:

pnpm ai-sdk -u <uuid> "Send a funny joke to the #random channel in Slack"
pnpm openai-sdk -u <uuid> "Send a funny joke to the #random channel in Slack"