Overview
What is Agent.ai MCP Server?
An MCP server implementation that integrates with the Agent.ai API, providing web text extraction, web screenshots, and YouTube transcript capabilities through a dynamic function loading system.
How to use Agent.ai MCP Server?
Configure the server with your Agent.ai API token, then run it via Docker or npx. Add the corresponding entry to your claude_desktop_config.json with the environment variable API_TOKEN set to your token.
Key features of Agent.ai MCP Server
- Dynamic function loading from the Agent.ai API
- Web text extraction (scrape or crawl up to 100 pages)
- Web screenshot capture with optional cache expiration
- YouTube video transcript retrieval
- Efficient caching of function definitions to reduce API calls
Use cases of Agent.ai MCP Server
- Extract text content from web pages for analysis or archiving
- Capture visual screenshots of websites for monitoring or documentation
- Fetch transcripts from YouTube videos for accessibility or content processing
- Crawl websites (up to 100 pages) to gather text data at scale
- Dynamically discover and call dozens of additional Agent.ai API tools
FAQ from Agent.ai MCP Server
What does the Agent.ai MCP Server do?
It acts as a bridge between MCP-compatible clients (like Claude Desktop) and the Agent.ai API, allowing you to call Agent.ai tools – such as web scraping, screenshots, and YouTube transcript extraction – directly from your AI assistant.
What are the prerequisites for running the server?
You need an Agent.ai API token and either Docker or Node.js (for npx). The server is configured via an API_TOKEN environment variable.
How do I obtain an Agent.ai API token?
The README states you must contact Agent.ai directly to obtain an API token. No further details are provided.
Where can I find the full list of available tools?
The README describes three tools (grab_web_text, grab_web_screenshot, get_youtube_transcript) and notes that dozens more are available dynamically. The complete list can be found at https://docs.agent.ai/api-reference.
Does the server cache any data?
Yes, function definitions are cached to reduce API calls. Additionally, the web screenshot tool supports a ttl_for_screenshot parameter to control cache expiry for screenshots.
