agentai

Agent.ai MCP Server exposes a dynamic, API-driven set of tools for web content extraction, screenshots, and YouTube transcripts. The server automatically discovers available functions from the Agent.ai API and exposes them as actions like grab_web_text, grab_web_screenshot, and get_youtube_transcript. To use the server, you must provide an API token via the API_TOKEN environment variable (or your deployment platform's secret management). Clients can then call the specific endpoints exposed by the MCP server to perform tasks such as extracting text from a page, capturing a screenshot, or retrieving a video transcript, all while leveraging caching to minimize repeated API calls.

The primary tools currently available include:

• grab_web_text: Extracts text from a web page. Required input: url. Optional mode can be scrape for a single page or crawl for up to 100 pages.
• grab_web_screenshot: Captures a screenshot of a web page. Required input: url. Optional ttl_for_screenshot sets cache expiration in seconds.
• get_youtube_transcript: Retrieves transcripts for a YouTube video. Required input: url.

These tools are accessed via the Agent.ai API endpoints under v1/action, such as /grab_web_text, /grab_web_screenshot, and /get_youtube_transcript. Example usage patterns and payloads are documented in the Agent.ai API reference linked in the README, and the MCP server will route requests accordingly while handling authentication and caching.

Prerequisites:

• Docker installed and running on your system
• An Agent.ai API token (you can request one from Agent.ai)

Step-by-step installation:

1. Ensure Docker is installed and running. Verify with:

docker --version

2. Obtain your Agent.ai API token and keep it ready. You will provide it to the MCP server via the API_TOKEN environment variable when starting the container.
3. Run the MCP server using Docker:

docker run -i --rm -e API_TOKEN=YOUR_API_TOKEN_HERE mcp/agentai

Replace YOUR_API_TOKEN_HERE with your actual token. The server should start and be ready to receive requests at the configured endpoints.

Optional alternative (NPX):

npx -y @agentai/mcp-server

This will pull and run the MCP server without using Docker, if you prefer a Node-based execution path.

Environment variables and configuration:

• API_TOKEN must be set to your Agent.ai API token for the MCP server to authenticate with Agent.ai.
• The Docker deployment uses the --rm flag to clean up the container after exit; remove it if you need the container to persist for debugging.
• The server caches function definitions to reduce API calls; you can adjust caching behavior through Agent.ai settings if available.
• If you encounter authentication errors, double-check that the API_TOKEN is correct and has the necessary permissions.
• When using grab_web_text in crawl mode, ensure the target sites permit crawling and respect robots.txt policies.
• For YouTube transcripts, ensure the provided video URL is accessible and not restricted by region or privacy settings.