apify

The Apify MCP Server lets AI agents execute Apify Actors as tools through the MCP interface. It enables OAuth-based authentication and supports connecting clients like Claude.ai, VS Code, and others via a hosted endpoint at mcp.apify.com. You can run Actors from the Apify Store to perform tasks such as scraping social media posts, extracting contact details from maps, or performing web searches, all through a consistent MCP workflow. The server supports both an HTTPS endpoint for hosted usage and a stdio workflow for local development and CLI-based clients, making it flexible for experimentation and production alike. When using the hosted service, you can leverage OAuth to connect from clients and use the Streamable HTTP transport, while local setups typically rely on the npx invocation with your API token.

To interact with the Apify MCP Server, connect using the MCP client of your choice and authorize via OAuth or provide your APIFY_TOKEN in the request headers. The server exposes Apify Actors as tools, and it includes built-in support for dynamic tool discovery and structured actor results. For local testing, run the server with the npx command and set APIFY_TOKEN in your environment, then configure your MCP client to point at the local or hosted endpoint as needed.

Prerequisites:

• Node.js (recommended LTS) and npm installed on your machine
• Internet access to fetch the MCP server package from npm

Installation steps:

1. Install or run via npx (no global install required):

   • npx -y @apify/actors-mcp-server

   Note: You can also install globally if you prefer:

   • npm install -g @apify/actors-mcp-server
   • Then run: actors-mcp-server

2. If you plan to run locally with stdio and provide an API token, set the token in the environment:

   • In Unix-like shells: export APIFY_TOKEN=your_apify_api_token
   • In Windows CMD: set APIFY_TOKEN=your_apify_api_token

3. Verify installation by checking help or version:

   • npx -y @apify/actors-mcp-server --help
   • or with a global install: actors-mcp-server --help

4. (Optional) For advanced usage, configure environment variables or a docker/uvx setup as per your deployment preferences.

Tips and caveats:

• The Apify MCP Server supports OAuth, enabling clients like Claude.ai and VS Code to connect via URL endpoints. If you’re using the hosted server, you’ll typically use the downstream OAuth flow rather than passing API tokens in every request.
• Starting with the hosted service, SSE transport is deprecated in favor of Streamable HTTP. Update your MCP client config accordingly to avoid disruptions after the deprecation date.
• When running locally via stdio, set APIFY_TOKEN to your Apify API token to authorize actor runs.
• If you rely on dynamic tool discovery features, ensure your MCP client supports them, and be aware some clients may require manual tool reloads for certain tool sets.
• The npm package name is @apify/actors-mcp-server; reference this in your deployment scripts or Docker invocations if you switch to containerized setups.
• For production deployments, consider securing your environment and limiting token exposure, and monitor the hosted service for any API or endpoint updates from Apify.