# How To Create Upgraded AI Agents with Gram

Source: https://www.youtube.com/watch?v=mkACzUDc8Vc
Recap page: https://rapidrecap.app/video/mkACzUDc8Vc
Generated: 2025-11-11T12:31:44.106+00:00

---
## Quick Overview

The video demonstrates how to integrate external APIs like Twitter Scraper and Instagram Scraper with Gram's Model Context Protocol (MCP) servers using OpenAPI specifications, creating custom tools that chain these APIs together to perform complex tasks like scraping YouTube video metadata and generating a summary via the Bumpups chat tool, all manageable through Gram's platform.

**Key Points:**
- The primary goal is to upgrade AI agents by connecting external APIs (like Tweet Scraper, TikTok Scraper, Instagram Scraper) to a central MCP layer using OpenAPI specifications.
- The demonstration focuses on integrating a YouTube Scraper API (Apify) with a Bumpups API tool to fetch video metadata and generate a summary in a single custom toolset.
- The process involves uploading the OpenAPI specification (YAML or JSON), naming the API (e.g., youtube_scraper), creating a toolset, and configuring the MCP server slug (e.g., webcafe-ai-youtube-grab).
- Environment variables (like the BUMPUPS_API_KEY_AUTH) must be set in the specified environment (e.g., youtube_data) for the tools within the toolset to function correctly.
- The resulting custom tool, named 'youtube_yes', chains the youtube_scraper_get_data tool with the bumpups_create_chat tool to achieve the composite goal of scraping metadata and summarizing content.
- The custom tool can then be tested in the Playground using the appropriate environment, successfully retrieving metadata (Title, Duration, Views, Likes, Comments) and a video summary based on GitHub best practices.

![Screenshot at 00:03: The central diagram illustrates the MCP concept, showing external scrapers like Tweet Scraper, TikTok Scraper, and Instagram Scraper feeding into the central 'MCP' layer, which an AI agent can then utilize.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-00-03.png)

**Context:** The video tutorial explains how to enhance AI agent capabilities within the Gram platform by leveraging the Model Context Protocol (MCP). MCP allows developers to easily connect various third-party APIs, referred to as 'scrapers' (like those for Twitter, TikTok, Instagram, and YouTube), into a unified, self-hosted server environment that AI models can utilize as tools. The core concept demonstrated is using OpenAPI specifications to rapidly generate these integrated tools without extensive manual coding.

## Detailed Analysis

The tutorial details the process of upgrading AI agents by integrating external APIs into Gram's MCP framework, starting by showing example scrapers for Twitter, TikTok, and Instagram that feed into the central MCP structure. The main demonstration involves creating a custom tool that chains the Apify YouTube Scraper with the Bumpups chat API. This is achieved by first uploading the OpenAPI specification (YAML or JSON) for the YouTube Scraper into Gram's 'Toolsets' section (00:48). Next, an MCP server is configured using a specific slug (02:54), and the necessary environment variable (BUMPUPS_API_KEY_AUTH) is set in a dedicated environment named 'youtube_data' (05:53). The creation of a custom tool, named 'youtube_yes' (13:13), involves chaining the youtube_scraper_get_data tool (for metadata) with the bumpups_create_chat tool (for summarization) (12:43). Once this composite tool is created and installed on the MCP server (09:49), testing it in the Playground confirms that the chained actions execute successfully, first running youtube_scraper_get_data to retrieve metadata (title, duration, views, likes, comments) and then using bumpups_create_chat to provide a summary of the video content, demonstrating seamless multi-API orchestration.

### API Integration Overview

- Diagram shows Tweet Scraper, TikTok Scraper, and Instagram Scraper feeding into MCP
- MCP allows connecting any API found online via OpenAPI spec
- Demonstrates connecting YouTube Scraper and Bumpups API.

### MCP Server Setup

- Upload OpenAPI spec (YAML/JSON) for YouTube Scraper (00:48)
- Name API (youtube_scraper)
- Configure MCP server slug (webcafe-ai-youtube-grab) (02:54)
- Set environment variables (BUMPUPS_API_KEY_AUTH) in 'youtube_data' environment (05:53).

### Custom Tool Creation (Toolify)

- Select youtube_scraper toolset
- Define custom tool 'youtube_yes' to chain youtube_scraper_get_data and bumpups_create_chat (12:43)
- Steps explicitly show tool chaining for metadata retrieval then summarization.

### Testing and Results

- Custom tool is added to the installed MCP server (09:49)
- Testing in Playground with the correct environment shows successful execution of both chained tools
- Results include YouTube metadata (Title: 'How To Use GitHub For Beginners') and a video summary based on GitHub branching best practices (12:23).

![Screenshot at 00:03: Initial diagram showing the concept of connecting various social media scrapers \(Twitter, TikTok, Instagram\) to a central MCP.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-00-03.png)
![Screenshot at 00:08: The Gram platform interface used to configure the YouTube Scraper MCP server, showing the command and environment variable setup.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-00-08.png)
![Screenshot at 00:49: The Apify Store interface where the YouTube Scraper tool is selected for integration into Gram.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-00-49.png)
![Screenshot at 01:29: Display of the YouTube Scraper's OpenAPI definition in JSON format, which Gram uses to generate tools.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-01-29.png)
![Screenshot at 02:06: The Cursr AI chat interface prompting the model to generate the OpenAPI specification as a JSON file.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-02-06.png)
![Screenshot at 04:51: The 'Add Tools to the toolset youtube\_scraper' dialog showing the selection of tools from the Bumpups API \(e.g., create\_chat, generate\_timestamps\).](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-04-51.png)
![Screenshot at 08:33: The installation screen for the custom tool, showing options to install via Cursor, Claude Code, Gemini CLI, etc.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-08-33.png)
![Screenshot at 11:50: The Cursr chat interface showing the planned execution sequence: running youtube\_scraper\_get\_data followed by bumpups\_create\_chat.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-11-50.png)
![Screenshot at 12:23: The final results pane showing the successfully retrieved YouTube metadata and the video summary generated by the custom chained tool.](https://ss.rapidrecap.app/screens/mkACzUDc8Vc/00-12-23.png)