What it is
Apify MCP Server lets an agent use any Apify Actor (a packaged scraper or automation) as a tool. By default it loads two Apify-built Actors, apify/rag-web-browser and apify/web-fetch, plus helper tools such as search-actors, fetch-actor-details and call-actor, so the agent can find a scraper for Instagram, Google Maps or a store page and run it. Version 0.19.3 shipped on 21 October 2026, and the repository has about 12,700 GitHub stars.
Connect to the hosted server at https://mcp.apify.com with OAuth or an Apify token, or run it locally with npx @apify/actors-mcp-server and an APIFY_TOKEN. The hosted version has features, such as output schema inference, that the local one lacks, and the old /sse endpoint has been removed. Agents without an Apify account can pay per run through x402, Skyfire or a prepaid AGI token.
Install
https://mcp.apify.comOur review
The strength is coverage. Instead of teaching an agent to scrape Instagram or Google Maps, you let it search Apify Store, read an Actor's input schema with fetch-actor-details and run it with call-actor. The default set is about a dozen tools, including two Apify-built Actors for browsing and page fetching, and the full catalogue runs to about 44, from schedules and builds to a raw apify-api-write tool that can POST or DELETE against any API path.
That breadth is billed. Every run spends platform credit at $0.13 to $0.20 per compute unit, and many Store Actors add their own per-result or per-event charges. The docs' answer is maxTotalChargeUsd in callOptions, which caps a single run, plus the account-wide usage limit; an agent looping through searches without either can burn the free plan's $5 before anyone looks. call-actor waits at most 45 seconds and returns storage IDs rather than data, so each scrape is at least two tool calls.
Security needs a careful read because the server runs other people's code on your behalf. In May 2026 Apify published two advisories: CVE-2026-50143, rated high, let a malicious Actor's metadata redirect requests and leak the caller's Apify token (fixed in 0.10.11), and CVE-2026-46341 let fetch-apify-docs fetch arbitrary sites through a prefix-matching flaw (fixed in 0.9.21). Both are patched, but the pattern matters more than the bugs: Actor definitions and scraped pages are untrusted input. The server marks most third-party Actor tools as destructive, and delete-actor loads only when named.
Telemetry is on by default for tool calls, and the stdio build also reports errors to Sentry; TELEMETRY_ENABLED=false turns both off. The README tells production users to pin an explicit tools list because the defaults may change between releases, which is honest and also a sign of how fast this surface moves. For reading arbitrary pages Firecrawl is simpler; Apify earns its place when the data sits behind a large platform someone else has already learned to scrape.
Where it earns its place
A growth marketer at a 15-person agency, working in Claude.ai
Pull Google Maps listings with contact details for a prospect list
Connect https://mcp.apify.com with OAuth, have the agent find a Maps scraper with search-actors and run it with a maxItems limit. The Free plan's $5 of monthly usage covers a test; regular pulls usually mean Starter at $19 plus the Actor's own per-result fees.
An AI engineer building a market-research agent on the Claude Agent SDK
Give the agent social and e-commerce data without writing scrapers
Pin the hosted URL to tools=actors plus a short list of named Actors, set maxTotalChargeUsd on every call and a hard usage limit on the account. Each scrape is call-actor then get-dataset-items, so budget two tool calls and up to 45 seconds per run.
A platform engineer who maintains the company's own Actors
Fix and redeploy a broken scraper from Claude Code
Load the versions, builds and runs categories: the agent reads files with get-actor-version, edits them with update-actor-version, calls build-actor and checks get-actor-run-log. That token can change production Actors, so keep it off shared machines.
Before you commit
- Setup
- Add https://mcp.apify.com to Claude.ai, Claude Code, Cursor or VS Code and sign in with OAuth, or send an Authorization: Bearer token. Locally, run npx @apify/actors-mcp-server with APIFY_TOKEN. Discovery and docs tools work without an account; running Actors needs a token or an agentic payment.
- Integrations
- About 12 tools by default (search-actors, fetch-actor-details, call-actor, four run and storage helpers, two docs tools, rag-web-browser and web-fetch) and about 44 in total across actors, runs, storage, tasks, schedules, builds, versions and raw API access. Any Store Actor can be added as its own tool by name.
- Data and privacy
- The token the server holds can run Actors, read every dataset and, with apify-api-write loaded, change anything in the account. Scraped pages and Actor READMEs are untrusted text in the model's context. Telemetry, and Sentry error reporting on stdio, are on by default. The platform has a SOC 2 Type II report, shared under NDA.
- Cost at scale
- The server is free; runs are billed at $0.13 to $0.20 per compute unit by plan, residential proxy at $7 to $8 per GB, plus per-result or per-event Actor fees. Plans are Free ($5 of usage), Starter $19, Scale $199 and Business $999. Accountless agents can prepay $1 by x402 or a $5 minimum Skyfire token.
- Limits
- Apify's API allows 250,000 requests a minute per user and 60 a second per resource; call-actor waits at most 45 seconds before handing back a run ID. The legacy /sse endpoint is gone, default tools may change between releases, and the repository has 232 open issues and pull requests.
What's good
- Thousands of Store Actors through one server
- Hosted endpoint with OAuth; local stdio via npx
- Per-call spend cap with maxTotalChargeUsd
- Destructive tools off by default; tool list set by URL parameter
What's not
- Run costs and third-party Actor fees add up quickly
- Two 2026 security advisories, one leaking API tokens
- Telemetry on by default
- Local stdio lacks hosted-only features
Best for
Agents that need structured data from specific platforms such as social networks, maps or online stores.
Alternatives
Apify MCP Server: common questions
- Is Apify MCP Server worth it?
- Apify's server puts the whole Apify Store within an agent's reach, which is both its strength and its bill; cap the spend per call before you let it loose. We rate it 4.1 out of 5.
- How much does Apify MCP Server cost?
- Server is free; Actor runs are billed. Free plan includes $5 of monthly usage; Starter $19/month, Scale $199/month, Business $999/month, each plus pay-as-you-go.
- What are the downsides of Apify MCP Server?
- Run costs and third-party Actor fees add up quickly; two 2026 security advisories, one leaking API tokens; telemetry on by default; local stdio lacks hosted-only features.
- Who is Apify MCP Server best for?
- Agents that need structured data from specific platforms such as social networks, maps or online stores.
- How much does Apify MCP Server cost for a team or in production?
- The server is free; runs are billed at $0.13 to $0.20 per compute unit by plan, residential proxy at $7 to $8 per GB, plus per-result or per-event Actor fees. Plans are Free ($5 of usage), Starter $19, Scale $199 and Business $999. Accountless agents can prepay $1 by x402 or a $5 minimum Skyfire token.
- Is Apify MCP Server safe for company data?
- The token the server holds can run Actors, read every dataset and, with apify-api-write loaded, change anything in the account. Scraped pages and Actor READMEs are untrusted text in the model's context. Telemetry, and Sentry error reporting on stdio, are on by default. The platform has a SOC 2 Type II report, shared under NDA.
- What are the best Apify MCP Server alternatives?
- The closest alternatives we have reviewed are Firecrawl MCP and Browserbase MCP.
Answers checked Oct 22, 2026.
Make Apify MCP Server? Show this rating on your site
