Apify MCP Server — 59,000+ Scrapers at Your AI Agent's Fingertips

Most MCP servers give your agent one capability — search, or database access, or browser automation. The Apify MCP Server takes a different approach: it gives your agent access to an entire marketplace of 59,000+ pre-built scrapers, crawlers, and automation tools (called “Actors” in Apify’s terminology), and lets the agent discover and use new ones on the fly.

Need Amazon product data? There’s an Actor for that. Google Maps leads? Actor. LinkedIn profiles, Instagram posts, TikTok videos, real estate listings? Actors for all of them. The MCP server is the bridge that lets your AI agent search this marketplace, understand what each tool expects, run it, and get structured results back.

What It Does

As of the current README, the Apify MCP Server exposes 19 tools organized around four capabilities: discovering Actors, running them, accessing stored results, and browsing Apify’s documentation.

Actor Discovery & Execution

Note: the earlier add-actor tool (which dynamically registered an Actor as a discrete new tool) was retired in v0.13.0 (July 2026) as part of a stateless-architecture migration. Clients that pass tools=add-actor now get it silently ignored; the server’s own docs say to use call-actor directly instead.

Run Management

Storage Access

Documentation & Dev

The dynamic tool discovery workflow is still the standout feature, though the exact mechanics changed with the July 2026 stateless migration. Rather than pre-configuring specific scrapers, your agent now:

1. Searches the Apify Store for relevant Actors using search-actors
2. Inspects Actor details to understand required inputs via fetch-actor-details
3. Calls the Actor directly with call-actor, passing the Actor ID and input — no separate “add as a tool” step

This is a simpler flow than the earlier add-actor model, per the current README: “It allows an AI agent to find new tools (Actors) as needed and incorporate them.” Your agent’s capabilities still grow dynamically based on what it needs, without manual reconfiguration — it just doesn’t register each discovered Actor as its own named tool anymore.

Hosted vs Local

Apify offers two ways to run the MCP server, and the differences matter:

The hosted mode at https://mcp.apify.com is the recommended path. It supports OAuth (no token exposure), infers output schemas so agents better understand what they’ll get back, and requires zero local setup. The Apify UI configurator even offers one-click install buttons for supported clients.

SSE deprecated: Server-Sent Events transport was removed on April 1, 2026 as planned, replaced by Streamable HTTP per the MCP specification. If you configured the SSE endpoint (/sse), update to the Streamable HTTP endpoint (/mcp).

Setup

Claude Desktop (hosted with OAuth):

Use the one-click install at mcp.apify.com, or add manually:

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_APIFY_TOKEN"
      }
    }
  }
}

(Note: The SSE endpoint /sse was removed April 1, 2026. Use /mcp for Streamable HTTP.)

Claude Desktop (local):

{
  "mcpServers": {
    "actors-mcp-server": {
      "command": "npx",
      "args": ["-y", "@apify/actors-mcp-server"],
      "env": {
        "APIFY_TOKEN": "YOUR_APIFY_TOKEN"
      }
    }
  }
}

Claude Code:

claude mcp add apify -- npx -y @apify/actors-mcp-server

Set APIFY_TOKEN in your environment.

Cursor:

Create .cursor/mcp.json pointing to the hosted URL with Bearer token authentication.

Pricing Reality

The MCP server itself is free and open source (MIT license — not Apache 2.0, correcting an earlier version of this review). But using it costs money because running Actors consumes Apify platform credits. Per Apify’s pricing page:

• Free tier: $5/month in credits (no credit card required). Enough for hundreds of small scraping tasks — testing, learning, small projects.
• Starter: $29/month
• Scale: $199/month
• Business: $999/month
• Enterprise: Custom pricing

Individual Actor pricing varies widely. Some community Actors are free; premium ones charge per result or per compute unit. Always use fetch-actor-details to check pricing before running an Actor.

The Skyfire integration enables autonomous agent payments via PAY tokens (minimum $5 USD), letting agents pay for Actor runs without pre-configured Apify tokens — interesting for fully autonomous agent workflows.

How It Compares

The Apify MCP Server competes in the web data extraction space. Here’s how it stacks up:

Star counts and licenses checked directly against each project’s GitHub repository on 2026-08-15; Firecrawl’s MCP server has moved to the firecrawl/firecrawl-mcp-server repo (previously under the mendableai org).

The key differentiator is breadth through marketplace: instead of one general-purpose scraper, you get thousands of purpose-built ones. The tradeoff is quality variance — community-created Actors range from excellent to abandoned, and you need to evaluate each one. Bright Data’s scraper library lists 1,000+ pre-built, officially maintained scrapers, while Apify’s 59,000+ Actors are community-curated and vary widely in quality.

Bright Data’s officially maintained scraper library offers higher consistency for enterprise deployments, while Apify’s marketplace model means there’s likely a specialized Actor for niche sites that general-purpose crawlers can’t handle.

What’s Good

• Unmatched breadth — 59,000+ Actors covering social media, search engines, e-commerce, maps, real estate, and more
• Dynamic tool discovery — agents find and call new scrapers at runtime without configuration changes
• Hosted mode is smooth — OAuth login, no tokens to manage, output schema inference
• Free tier — $5/month in credits with no credit card required, enough for testing and small projects
• Active development — v0.14.3 released August 12, 2026; the server shipped a breaking stateless-architecture migration across v0.12.0–v0.14.0 (July 2026), added abort-actor-run and report-problem tools, and retired the older add-actor tool
• MIT license — permissive open-source license for the server itself (corrected from an earlier version of this review, which stated Apache 2.0)

What’s Not

• Platform dependency — this is an MCP gateway to a commercial SaaS platform, not a standalone tool. You’re dependent on Apify’s infrastructure, pricing, and continued operation
• Actor quality is uneven — community-created scrapers vary widely in reliability, maintenance, and accuracy. Some are excellent; others are abandoned with broken selectors
• Costs can surprise you — running premium Actors or high-volume scraping burns through credits quickly. A few careless call-actor invocations by an autonomous agent could drain your free tier. The paid tiers above Starter also jump quickly (Scale is $199/month, not the $39/month this review previously stated)
• Rate limiting — 30 requests/second per user. Heavy autonomous agent loops can hit this
• Telemetry on by default — both Sentry error tracking and usage telemetry are enabled unless you opt out
• Workflow churn — the July 2026 stateless migration removed the add-actor tool with no deprecation window beyond one minor version, meaning integrations built around registering Actors as named tools broke and had to be rewritten around call-actor

The Bottom Line

The Apify MCP Server is the most capable web data extraction gateway available through MCP. Its dynamic tool discovery model — where agents can search a marketplace of 59,000+ scrapers and call them as tools at runtime — is unique and genuinely powerful. No other MCP server offers this level of adaptive capability.

But it’s a gateway to a commercial platform, not a self-contained tool. You’re paying Apify for every Actor run, quality varies across community-created scrapers, and the server’s rapid breaking-change pace (four minor versions in July 2026 alone, including a tool removal) is worth planning around. For small-to-medium scraping needs with the free tier, it’s an excellent way to give agents structured web data. For enterprise-scale extraction, Bright Data’s officially maintained scraper library and higher reliability may justify the premium.

Rating: 3.5 / 5 — Unique marketplace model with dynamic discovery, held back by platform dependency, uneven Actor quality, and a fast-moving API surface that has already broken at least one advertised workflow (add-actor). The free tier makes it worth trying; just watch your credit consumption with autonomous agents.

This review was originally published March 23, 2026, refreshed May 3, 2026, and claim-level audited August 15, 2026. Audit findings: latest release is v0.14.3 (August 12, 2026), not v0.9.16 — the server went through a breaking stateless-architecture migration (v0.12.0–v0.14.0, July 2026) that retired the add-actor tool and added abort-actor-run/report-problem. License is MIT, not Apache 2.0. Stars grew to ~3,700+. The Apify Store now lists 59,000+ Actors, not 3,000+. Apify’s paid “Scale” plan is $199/month, not $39/month. Node.js requirement is now v22+. Competitor star counts and licenses (Firecrawl, Bright Data, Crawl4AI) were re-verified directly against their GitHub repos. The previous “18.8% stress-test success rate” claim was removed — its sole source is a benchmark that discloses a sponsorship relationship with Apify, and no independent second source could be found. We research MCP servers thoroughly but do not test them hands-on. Check the GitHub repository for the latest updates.