Apify Claude Connector: The Complete Guide

The Apify MCP server enables AI agents to extract data from social media, search engines, maps, e-commerce sites or any other website using thousands of ready-made scrapers, crawlers and automation tools available on the Apify Store. The official server is hosted at mcp.apify.com with source at github.com/apify/apify-mcp-server.

Out of the box the hosted server loads a deliberately lean default toolset rather than the whole Store: search-actors to find Actors in the Apify Store, fetch-actor-details to inspect an Actor's inputs and pricing, the general-purpose apify/rag-web-browser for browsing and extracting web data, search-apify-docs and fetch-apify-docs for documentation lookups, and get-actor-output to retrieve results from a run. On clients that support dynamic tool discovery, the agent also gets an add-actor tool - so Claude can search the Store mid-conversation, inspect an Actor's required inputs, bolt it on as a new tool and execute it without any reconfiguration. You can customise the exposed surface with a query parameter (for example https://mcp.apify.com?tools=actors,docs,apify/rag-web-browser), and opt out of telemetry with ?telemetry-enabled=false.

The clever architectural choice: Actors are MCP tools. Each pre-built scraper or automation in the Apify Store becomes a tool Claude can discover, parameterise and call. No custom coding required - the AI triggers real web-data extraction, processing and automation on Apify's cloud infrastructure.

That turns the Apify Store (5,000+ Actors and growing) into a marketplace of capabilities the agent can browse, which is qualitatively different from connectors that expose a fixed surface.

• Scrape any modern website - via the Web Scraper MCP or category-specific Actors.
• Social media extraction - Facebook Posts Scraper, Instagram Scraper for posts, profiles, places, photos, comments.
• Maps and contacts - Google Maps Email Extractor.
• Search engines - Google Search Results Scraper.
• E-commerce - product, pricing and listing scrapers across major platforms.
• Documentation indexing - turn entire doc sites into machine-readable corpora.

• Competitor pricing intelligence: have Claude scrape rivals' pricing pages and produce a comparison table.
• Documentation comparisons: extract and compare docs across vendors as part of a build-vs-buy review.
• Lead generation: Google Maps Email Extractor surfaces contact information for a specific niche.
• RAG indexing: scrape entire documentation sites and feed the markdown to your embedding pipeline.
• Sentiment monitoring: social-media Actors return raw content for Claude to summarise.
• Search-engine SERP analysis: Google Search Results Scraper feeds SEO and brand-monitoring workflows.

One of the most powerful patterns the connector enables: feeding scraped markdown to embedding models and vector databases for retrieval. You point Claude at an Apify Actor that crawls a documentation site, pipe the output to your embedding pipeline, and now you have grounded, up-to-date answers in your agent without standing up a custom crawler.

For teams building RAG over external knowledge, this is one of the cleanest pipelines you can wire up.

There is no separate MCP fee. You pay normal Apify usage for Actor runs - compute, memory and per-Actor pricing as published. The free plan includes monthly credits, which is enough to test MCP with small runs.

For production volume, model it carefully - a recurring scrape on a large site can run real costs. The MCP doesn't change that economics, it just makes triggering the work easier.

Reception across Apify's own product pages, the apify-mcp-server GitHub repo, Use-Apify's beginner walkthroughs and setup guides, a2a-mcp.org's "Connect Claude and AI Agents to 5,000+ Web Scrapers" overview and Composio's Cowork integration is consistent: the connector turns "I need to scrape X" from a multi-day engineering task into a chat message.

Honest counterpoint: Actors vary in quality. Test the specific one you plan to depend on before wiring it into a production loop.

There are two routes. The hosted remote server is the recommended path for 2026 - it auto-updates, keeps no token in a local file and authenticates via OAuth. The local stdio server gives you a pinned, scriptable install with an explicit Actor whitelist.

Option A: Hosted remote server (recommended)

The official endpoint is https://mcp.apify.com, served over Streamable HTTP. (Note: the legacy SSE transport was removed on 1 April 2026 - if a guide tells you to use an /sse URL, it is out of date.) In Claude Code:

claude mcp add --transport http apify https://mcp.apify.com

On first use Claude opens an OAuth flow; sign in with your Apify account and approve. In Claude Desktop or Claude.ai, add it from Settings → Connectors instead - no token is stored locally, and Apify pushes server updates automatically.

Option B: Local stdio server

Requires Node.js v18+ on your PATH and an API token from the Apify Console (Settings → API & Integrations). Use the --actors flag to whitelist exactly the Store Actors you want exposed, which keeps the tool list small and the agent focused:

{
  "mcpServers": {
    "apify": {
      "command": "npx",
      "args": [
        "-y", "@apify/actors-mcp-server",
        "--actors", "apify/rag-web-browser,apify/instagram-scraper"
      ],
      "env": { "APIFY_TOKEN": "your_apify_token" }
    }
  }
}

Then browse the Apify Store for the Actor you need, and ask Claude to call it with your parameters.

• Still using an /sse URL: the SSE transport was retired on 1 April 2026. Switch to the Streamable HTTP endpoint https://mcp.apify.com (no /sse suffix).
• "command not found: npx" on the local server: Node.js isn't on your PATH. Install Node 18+ and confirm with node --version, or just use the hosted remote server, which needs no local runtime.
• Claude can't see your Actor: on the local server the --actors whitelist is exact - if the Actor isn't listed, it isn't exposed. Add its username/actor-name slug and restart.
• Run fails with "insufficient credits": Actor runs draw on your normal Apify balance. The free plan's monthly credits cover testing; heavy or recurring scrapes need a paid plan.
• Too many tools overwhelm the model: exposing the whole Store floods the context with tool definitions. Whitelist a handful of Actors (local) or ask Claude to use a specific one by name (remote).
• Scraper returns empty or blocked results: some sites are heavily JS-rendered or anti-bot-protected. Test the specific Actor in the Apify Console first, and prefer Actors with proxy/anti-blocking support for tough targets.
• OAuth loop on the remote server: clear the cached connector auth and re-add it; make sure you're signing into the same Apify account that holds your credits.

Auth is OAuth via Apify - the agent operates with your Apify account access. The bigger consideration is scraping responsibility: legal and ethical compliance for what you scrape is yours, not the connector's. Respect each site's terms of service, robots.txt and applicable laws. The connector makes the engineering trivial; the policy work is unchanged.

• Actor quality varies: test the specific Actor before depending on it.
• Real cost: heavy scraping uses real credits.
• Legal responsibility: scraping compliance is yours.
• Dynamic sites: some heavily JS-rendered or anti-bot-protected sites are harder.
• Rate-limited: the hosted server allows up to 30 requests per second per user - generous for chat, but a ceiling for aggressive agent loops.
• Local server is less capable: per Apify's docs, the local stdio server does not support the output-schema inference the hosted version has - another reason to prefer mcp.apify.com.

Data engineers, growth and research teams, anyone building RAG over external data, anyone running competitive intelligence or lead-gen pipelines. The "5,000+ scrapers as MCP tools" framing makes Apify uniquely useful for agents that need broad web reach.

The Apify connector turns the web into a callable tool surface. Thousands of Actors as MCP tools, OAuth auth, normal Apify pricing - and Claude as the orchestrator. For anyone whose work depends on getting external web data into an agent loop, it is one of the highest-leverage connectors available.

Pilot a small Actor, model the cost on your real volume, and respect each target site's terms. Browse more in the complete Claude Connectors Directory.