What Apify's MCP actually does

Apify MCP is Apify's bridge from an agent to Apify Store. The agent can search for Actors, inspect input and output shapes, run them, and read datasets or key-value stores. You get Instagram scrapers, Google Maps extractors, web fetch, and thousands of other ready-made jobs without writing a separate MCP server for each site.

This week it is still moving. Appsgit's MCP trending list for 30 Oct to 6 Nov 2026 ranks Apify MCP Server among the weekly star gainers. On PlugMCP the listing is under remote web scraping, publisher apify, transport http, status remote. The publisher endpoint is https://mcp.apify.com.

What it is good for

Use it when the agent needs structured data from the public web and you would rather run a maintained Actor than maintain a local scraper. Search Actors by use case, fetch details (including inferred output schemas on the hosted server), call an Actor, then pull items from the resulting dataset.

Default tools cover Actor search and run, Apify docs search, apify/rag-web-browser, apify/web-fetch, and report-problem. You can narrow the tool list with a tools query on the URL, or pin specific Actors so the agent only sees those.

The npm package for the local stdio path is @apify/actors-mcp-server (checked at 0.21.4 when this draft was written). The hosted path is the one Apify recommends for Cursor and Claude. Output schema inference for Actor results is a hosted feature. The local stdio server does not include it.

Install in Cursor

Preferred path from the publisher: open https://mcp.apify.com, pick Cursor, choose tools, and copy the config, or run:

1apify mcp install cursor

That writes the client config, including auth. Manual remote config that matches the PlugMCP listing:

1{
2  "mcpServers": {
3    "apify-apify-mcp-server": {
4      "url": "https://mcp.apify.com/"
5    }
6  }
7}

On first use the client opens Apify sign-in (OAuth). If your client cannot do OAuth, send a Bearer token instead:

1{
2  "mcpServers": {
3    "apify": {
4      "url": "https://mcp.apify.com",
5      "headers": {
6        "Authorization": "Bearer YOUR_APIFY_TOKEN"
7      }
8    }
9  }
10}

Treat the token like a password. Keep it out of git.

Local stdio when a remote URL is not an option (Node.js 22 or later per the README; Apify docs also mention Node 18 or higher for the docs path):

1{
2  "mcpServers": {
3    "actors-mcp-server": {
4      "command": "npx",
5      "args": ["-y", "@apify/actors-mcp-server"],
6      "env": {
7        "APIFY_TOKEN": "YOUR_APIFY_TOKEN"
8      }
9    }
10  }
11}

After it connects, ask the agent to search Apify Store for a simple scraper and show the input fields. You should see tool results, not an empty tool list.

Install in Claude

Claude Code:

1claude mcp add apify "https://mcp.apify.com" -t http

Or install the Apify plugin (MCP plus agent skills) from the publisher docs. Claude.ai and Claude Desktop can use the Apify connector, or a custom connector pointed at https://mcp.apify.com.

ChatGPT supports the same hosted URL as a custom MCP plugin with OAuth. No token needed for that path.

Limits to know

Running Actors and reading storage needs an Apify account (or an agentic prepaid token). Discovery-only tools can run without a token when you restrict the tools query to search and docs helpers.

call-actor returns run status and storage IDs, not the dataset rows. Read results with get-dataset-items. Long runs may return a run ID for polling after the wait window (up to 45 seconds, default 30).

Full-permission Actors are excluded from search and run so the agent cannot approve elevated access on your behalf.

Telemetry of tool names and client metadata is on by default. Opt out with telemetry-enabled=false on the hosted URL, or --telemetry-enabled=false / TELEMETRY_ENABLED=false for stdio.

Legacy SSE at https://mcp.apify.com/sse is gone. Use Streamable HTTP at https://mcp.apify.com.

Alternatives on PlugMCP

Same shelf, different jobs:

• Firecrawl when you want a remote scrape and search API without Apify Store Actors.
• Scrapling when the work should stay on your machine in Python.
• Playwright MCP or Chrome DevTools MCP when the job is interactive browser automation rather than Store scrapers.
• Browser Use when you want a managed browser-agent cloud.

On the PlugMCP page

The listing is apify-apify-mcp-server. Transport is http. Status is remote. The install block points at the publisher URL https://mcp.apify.com/. There is a second listing under com.apify-apify-mcp-server with the same remote URL.

Search more scraping and remote servers from the directory MCP at https://plugmcp.ai/mcp-server, or open the web scraping category on the site.