The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Actors MCP Server listing page.
The Apify Model Context Protocol (MCP) server at mcp.apify.com enables your AI agents to extract data from social media, search engines, maps, e-commerce sites, and any other website using thousands of ready-made scrapers, crawlers, and automation tools from Apify Store. It supports OAuth, allowing you to connect from clients like Claude.ai or Visual Studio Code using just the URL.
🚀 Use the hosted Apify MCP Server!
For the best experience, connect your AI assistant to our hosted server at https://mcp.apify.com. The hosted server supports the latest features - including output schema inference for structured Actor results - that are not available when running locally via stdio.
⚠️ Legacy SSE transport removed. The https://mcp.apify.com/sse endpoint has been removed in favor of Streamable HTTP. Migrate your client to https://mcp.apify.com — drop the /sse suffix from your configuration.
💰 The server also supports agentic payments: buy a token from AGI to run any Actor, or pay per-request via direct x402 (Pay Per Event Actors only) or Skyfire.
Apify MCP Server is compatible with Claude Code, Claude.ai, Cursor, VS Code and any client that adheres to the Model Context Protocol.
Check out the MCP clients section for more details or visit the MCP configuration page.
Use Instagram Scraper to scrape Instagram posts, profiles, places, photos, and comments.
Use RAG Web Browser to search the web, scrape the top N URLs, and return their content.
Use Web Fetch to fetch any URL and return its content as Markdown, plain text, HTML, or links — with JavaScript rendering and anti-bot protection.
Video tutorial: Integrate 8,000+ Apify Actors and Agents with Claude
🚀 Quickstart
You can use the Apify MCP Server in two ways:
HTTPS Endpoint (mcp.apify.com): Connect from your MCP client via OAuth or by including the Authorization: Bearer <APIFY_TOKEN> header in your requests. This is the recommended method for most use cases. Because it supports OAuth, you can connect from clients like Claude.ai or Visual Studio Code using just the URL: https://mcp.apify.com.
https://mcp.apify.com streamable transport
Standard Input/Output (stdio): Ideal for local integrations and command-line tools like the Claude for Desktop client.
Set the MCP client server command to npx @apify/actors-mcp-server and the APIFY_TOKEN environment variable to your Apify API token.
See npx @apify/actors-mcp-server --help for more options.
You can find detailed instructions for setting up the MCP server in the Apify documentation.
🤖 MCP clients
Apify MCP Server is compatible with any MCP client that adheres to the Model Context Protocol, but the level of support for dynamic tool discovery and other features may vary between clients.
This interactive, chat-like interface provides an easy way to explore the capabilities of Apify MCP without any local setup.
Sign in with your Apify account and start experimenting with web scraping, data extraction, and automation tools!
Or use the MCP bundle file (formerly known as Anthropic Desktop extension file, or DXT) for one-click installation: Apify MCP Server MCPB file
💰 Agentic payments
You can pay for Actor runs without an Apify API token using AGI, direct x402, or Skyfire.
AGI (agi.apify.com) mints a prepaid Apify API token in exchange for an x402 or MPP payment. Use the token like a normal API token against mcp.apify.com and api.apify.com — works for any Actor, not just Pay Per Event ones. Recommended for new integrations; see AGI (recommended) below.
Direct x402 pays with USDC on Base per request and does not require a separate platform account. It is fully supported by mcpc (brew install apify/tap/mcpc or npm install -g @apify/mcpc). We use mcpc because it is one of the few MCP clients that supports the latest features and the x402 protocol natively.
Skyfire pays with PAY tokens and requires a Skyfire account with a funded wallet. It does not require a special MCP client; the entire payment flow is handled directly through the MCP tool call parameters.
ℹ️ Scope: Both direct x402 and Skyfire are limited to Pay Per Event Actors, don't support Standby Actors, and settle per run instead of minting a token.
How agentic payments work
Actor run costs vary, so both payment methods use a prepaid balance model. The payment flow happens in four steps:
Discovery: The agent discovers Actors with search-actors or fetch-actor-details. Those calls are free.
Prepayment: Before running a paid Actor tool, the agent funds a prepaid balance.
Direct x402: mcpc automatically signs a $1.00 USDC transaction.
Skyfire: The agent creates a PAY token (minimum $5.00) using Skyfire's create-pay-token tool.
Execution: The agent calls the Actor tool.
Direct x402: Handled automatically by mcpc using the prepaid balance.
Skyfire: The agent explicitly passes the PAY token in the skyfire-pay-id input property.
Resolution: The tool returns the Actor results. Unused funds stay available for later runs.
Direct x402: After 60 minutes of inactivity, the server refunds any unused balance to the wallet on Base.
Skyfire: Skyfire returns unused funds when the token expires.
🪙 AGI (recommended)
AGI (Apify Agent General Interface) is the recommended way for autonomous agents to pay for Apify usage without an account. Pay once via x402 or MPP, receive a prepaid, spend-capped Apify API token, and use it directly against mcp.apify.com and api.apify.com (Authorization: Bearer <token>) — for any Actor.
Full protocol, supported payment methods, and current terms (minimum amount, token lifetime, refund policy) are documented at agi.apify.com/AGENTS.md — treat it as the single source of truth.
💸 Direct x402
The x402 protocol enables direct, machine-to-machine payments. Your MCP client can use it to pay for Actor runs with USDC on the Base blockchain, completely bypassing the need for an Apify API token.
# Create a new wallet
mcpc x402 init
# Import an existing wallet
mcpc x402 import <private-key>
# Show the wallet address and a funding QR code, so you can fund it with USDC on Base (https://base.org)
mcpc x402
mcpc @apify tools-call call-actor actor:="apify/rag-web-browser" input:='{"query": "latest AI news"}'
🔥 Skyfire
Skyfire provides managed payment infrastructure for AI agents. Instead of authenticating with an Apify API token, your agent passes a Skyfire payment token to cover the cost of each tool call using PAY tokens.
The MCP server provides a set of tools for interacting with Apify Actors.
Since Apify Store is large and growing rapidly, the MCP server provides a way to dynamically discover and use new Actors.
Actors
Any Apify Actor can be used as a tool.
By default, the server is pre-configured with two Actors, apify/rag-web-browser and apify/web-fetch, and several helper tools.
The MCP server loads an Actor's input schema and creates a corresponding MCP tool.
This allows the AI agent to know exactly what arguments to pass to the Actor and what to expect in return.
For example, for the apify/rag-web-browser Actor, the input parameters are:
config.json
{
"query": "restaurants in San Francisco",
"maxResults": 3
}
You don't need to manually specify which Actor to call or its input parameters; the LLM handles this automatically.
When a tool is called, the arguments are automatically passed to the Actor by the LLM.
You can refer to the specific Actor's documentation for a list of available arguments.
Helper tools
One of the most powerful features of using MCP with Apify is dynamic tool discovery.
It allows an AI agent to find new tools (Actors) as needed and incorporate them.
Here are some special MCP operations and how the Apify MCP Server supports them:
Apify Actors: Search for Actors, view their details, and use them as tools for the AI.
Apify documentation: Search the Apify documentation and fetch specific documents to provide context to the AI.
Actor runs: Get lists of your Actor runs, inspect their details, and retrieve logs.
Apify storage: Access data from your datasets and key-value stores.
Actor tasks: Create, inspect, and update your saved Actor tasks, and publish or unpublish their public landing pages.
Overview of available tools
Here is an overview list of all the tools provided by the Apify MCP Server.
Legend for the Enabled by default column:
✅ — in the default tool set.
⚡ — auto-injected when call-actor, an Actor tool, or get-actor-run is present (which is true in the default configuration).
✅¹ — served by default, but only when telemetry is enabled and the client is not withheld: Anthropic surfaces (Claude.ai / Claude Desktop / Claude Code) or local-agent-mode-apify. To disable, pass an explicit tools= list that omits it. Explicitly selecting it (tools=report-problem or tools=dev) serves it regardless of client — telemetry still must be enabled.
Tool name
Category
Description
Enabled by default
search-actors
actors
Search for Actors in Apify Store.
✅
fetch-actor-details
actors
Retrieve detailed information about a specific Actor, including its input schema, README (summary when available, full otherwise), pricing, and Actor output schema.
✅
call-actor
actors
Call an Actor and get its run results. Use fetch-actor-details first to get the Actor's input schema.
✅
get-actor-run
runs
Get detailed information about a specific Actor run.
⚡
get-dataset-items
storage
Retrieve items from a dataset with support for filtering and pagination.
⚡
get-key-value-store-record
storage
Get the value associated with a specific key in a key-value store.
⚡
abort-actor-run
runs
Abort a running Actor run, optionally gracefully.
⚡
search-apify-docs
docs
Search the Apify documentation for relevant pages.
✅
fetch-apify-docs
docs
Fetch the full content of an Apify documentation page by its URL.
An Actor tool to fetch a URL and return its content.
✅
report-problem
dev
Report a problem with an Apify tool or Actor to the Apify team.
✅¹
get-actor-run-list
runs
Get a list of an Actor's runs, filterable by status.
get-actor-log
runs
Retrieve the logs for a specific Actor run.
get-dataset
storage
Get metadata about a specific dataset.
get-dataset-schema
storage
Generate a JSON schema from dataset items.
get-key-value-store
storage
Get metadata about a specific key-value store.
get-key-value-store-keys
storage
List the keys within a specific key-value store.
get-dataset-list
storage
List all available datasets for the user.
get-key-value-store-list
storage
List all available key-value stores for the user.
create-actor-task
tasks
Create a saved Actor task (a named, reusable Actor configuration).
get-actor-task
tasks
Get a saved Actor task, its publication state and public display configuration.
update-actor-task
tasks
Update a task's input, run options, or public display configuration.
publish-actor-task
tasks
Publish a task on its public landing page.
unpublish-actor-task
tasks
Unpublish a task from its public landing page.
Note:
When call-actor, an Actor tool, or get-actor-run is present, the server auto-injects get-actor-run, get-dataset-items, get-key-value-store-record, and abort-actor-run.
When you call an Actor — through call-actor or directly via an Actor tool (e.g., apify--rag-web-browser) — the response contains run metadata, storage IDs, and a summary + nextStep, but no dataset items. To fetch items, follow nextStep and call get-dataset-items (auto-injected), passing the datasetId returned from the call.
Tool annotations
All tools include metadata annotations to help MCP clients and LLMs understand tool behavior:
title: Short display name for the tool (e.g., "Search Actors", "Call Actor", "apify/rag-web-browser")
readOnlyHint: true for tools that only read data without modifying state (e.g., get-dataset, fetch-actor-details)
openWorldHint: true for tools that access external resources outside the Apify platform (e.g., call-actor executes external Actors). Tools that interact only with the Apify platform (like search-actors or fetch-apify-docs) do not have this hint.
Tools configuration
The tools configuration parameter is used to specify loaded tools – either categories or specific tools directly, and Apify Actors. For example, tools=storage,runs loads two categories; tools=call-actor loads just one tool.
When no query parameters are provided, the MCP server loads the following tools by default:
actors
docs
apify/rag-web-browser
apify/web-fetch
If the tools parameter is specified, only the listed tools or categories will be enabled – no default tools will be included.
report-problem is served by default (subject to the gating in the footnote above) but lives in the dev category, so an explicit tools=dev selects it too. To disable it, pass an explicit tools= list that omits it (e.g. tools=actors,docs).
Easy configuration:
Use the UI configurator to configure your server, then copy the configuration to your client.
Configuring the hosted server:
The hosted server can be configured using query parameters in the URL. For example, to load the default tools, use:
For minimal configuration, if you want to use only a single Actor tool - without any discovery or generic calling tools, the server can be configured as follows:
Code
https://mcp.apify.com?tools=apify/my-actor
This setup exposes only the specified Actor (apify/my-actor) as a tool. No other tools will be available.
Configuring the CLI:
The CLI can be configured using command-line flags. For example, to load the same tools as in the hosted server configuration, use:
As above, this exposes only the specified Actor (apify/my-actor) as a tool. No other tools will be available.
⚠️ Important recommendation
The default tools configuration may change in future versions. When no tools parameter is specified, the server currently loads default tools, but this behavior is subject to change.
For production use and stable interfaces, always explicitly specify the tools parameter to ensure your configuration remains consistent across updates.
UI mode configuration
The ui parameter enables MCP Apps widget rendering in tool responses. When enabled, tools like search-actors return interactive MCP App responses.
Configuring the hosted server:
Enable UI mode using the ui query parameter:
Code
https://mcp.apify.com?ui=true
You can combine it with other parameters:
Code
https://mcp.apify.com?tools=actors,docs&ui=true
Configuring the CLI:
The CLI can be configured using command-line flags. For example, to enable UI mode:
Terminal
npx @apify/actors-mcp-server --ui true
You can also set it via the UI_MODE environment variable:
server.ts
export UI_MODE=true
npx @apify/actors-mcp-server
Backward compatibility
The v2 configuration preserves backward compatibility with v1 usage. Notes:
actors param (URL) and --actors flag (CLI) are still supported.
enableAddingActors (URL), enable-adding-actors (CLI), and the legacy enableActorAutoLoading alias have been removed. To call Actors dynamically, use tools=call-actor (included by default via the actors category). Any lingering raw value is ignored.
Defaults remain compatible: when no tools are specified, the server loads actors, docs, apify/rag-web-browser, and apify/web-fetch.
If any tools are specified, the defaults are not added (same as v1 intent for explicit selection).
call-actor is now included by default via the actors category (additive change). To exclude it, specify an explicit tools list without actors.
tools=add-actor, tools=experimental, and tools=preview are retired: they are ignored and load no tools. Use tools=call-actor (or the default actors category) instead.
Existing URLs and commands using ?actors=... or --actors continue to work unchanged.
Prompts
The server advertises the prompts capability, but no prompts are currently registered — prompts/list returns an empty list.
Resources
Your Apify data is not enumerated in resources/list — reads are on demand: pass any Apify API GET URL (https://api.apify.com/v2/...) to resources/read and the server injects the session's Apify token and returns the response body. resources/templates/list enumerates the common shapes — dataset items, key-value store records and keys, run metadata, run log — with their paging parameters. Responses inline up to 256 KB; anything larger returns a short notice with a download URL instead of the body. API reads require an Apify token, so a payment-only session (x402 or Skyfire) gets a JSON-RPC error for them.
💬 Usage examples
Below are realistic examples showing how an AI assistant uses the Apify MCP Server tools.
Example 1: Search the web using RAG Web Browser
User prompt:
Find the latest news about autonomous AI agents and summarize the key developments.
The AI assistant calls the pre-configured apify--rag-web-browser Actor tool to search the web and return content from top results.
The tool returns markdown content from the top 3 search results, which the AI assistant then summarizes for the user.
Example 2: Discover and run an Actor from Apify Store
User prompt:
Scrape the top 10 restaurants in Prague from Google Maps with their contact details.
The AI assistant first searches for a suitable Actor, inspects its input schema, and then executes it.
The tool returns a preview of the scraped data including restaurant names, addresses, ratings, phone numbers, and websites.
Example 3: Retrieve and paginate through Actor run results
User prompt:
Show me the next 10 results from that scraping run.
The AI assistant uses the dataset ID from the previous Actor run to fetch additional items.
Expected output: The tool returns the next page of structured data items from the Actor's output dataset.
📡 Telemetry
The Apify MCP Server collects telemetry data about tool calls to help Apify understand usage patterns and improve the service.
By default, telemetry is enabled for all tool calls.
The stdio transport also uses Sentry for error tracking, which helps us identify and fix issues faster.
Sentry is automatically disabled when telemetry is opted out.
Opting out of telemetry
You can opt out of telemetry (including Sentry error tracking) by setting the --telemetry-enabled CLI flag to false or the TELEMETRY_ENABLED environment variable to false.
CLI flags take precedence over environment variables.
Examples
For the remote server (mcp.apify.com):
text
# Disable via URL parameter
https://mcp.apify.com?telemetry-enabled=false
For the local stdio server:
server.ts
# Disable via CLI flag
npx @apify/actors-mcp-server --telemetry-enabled=false# Or set environment variable
export TELEMETRY_ENABLED=false
npx @apify/actors-mcp-server
⚙️ Development
Please see the CONTRIBUTING.md guide for contribution guidelines and commit message conventions.
For detailed development setup, project structure, and local testing instructions, see the DEVELOPMENT.md guide.
Upon launching, the Inspector will display a URL that you can open in your browser to begin debugging.
Unauthenticated access
When the tools query parameter includes only tools explicitly enabled for unauthenticated use, the hosted server allows access without an API token.
Currently allowed tools: search-actors, fetch-actor-details, search-apify-docs, fetch-apify-docs.
Example: https://mcp.apify.com?tools=search-actors.
🐦 Canary PR releases
Apify MCP is split across two repositories: this repository for core MCP logic and the private apify-mcp-server-internal for the hosted server.
Changes must be synchronized between both.
To create a canary release, add the beta label to your pull request.
This publishes the package to pkg.pr.new for staging and testing before merging.
See the workflow file for details.
🐋 Docker Hub integration
The Apify MCP Server is also available on Docker Hub, registered via the mcp-registry repository. The entry in servers/apify-mcp-server/server.yaml should be deployed automatically by the Docker Hub MCP registry (deployment frequency is unknown). Before making major changes to the stdio server version, test it locally to ensure the Docker build passes. To test, change the source.branch to your PR branch and run task build -- apify-mcp-server. For more details, see CONTRIBUTING.md.
The Actor input schema is processed to be compatible with most MCP clients while adhering to JSON Schema standards. The processing includes:
Descriptions longer than 500 characters (as defined in ACTOR_MAX_DESCRIPTION_LENGTH) are cut at the last complete sentence in that window, or the last complete word if there is no sentence break, and marked [Description truncated].
Enum fields that fit within 2000 combined characters (as defined in ACTOR_ENUM_MAX_LENGTH) are kept in full; larger enums are omitted, with a short note and a few example values in the description instead.
Required fields are explicitly marked with a REQUIRED prefix in their descriptions for compatibility with frameworks that may not handle the JSON schema properly.
Nested properties are built for special cases like proxy configuration and request list sources to ensure the correct input structure.
Array item types are inferred when not explicitly defined in the schema, using a priority order: explicit type in items > prefill type > default value type > editor type.
Enum values and examples are added to property descriptions to ensure visibility, even if the client doesn't fully support the JSON schema.
Rental Actors are only available for use with the hosted MCP server at https://mcp.apify.com. When running the server locally via stdio, you can only access Actors that are already added to your local toolset. To dynamically search for and use any Actor from Apify Store—including rental Actors—connect to the hosted endpoint.
🔒 Privacy policy
When you use this server, your requests and Actor inputs are sent to the Apify API for execution.
Data is not shared with third parties beyond what is necessary to run the requested Actors.
For full details on data collection, usage, sharing, and retention, see Apify Legal.
🤝 Contributing
We welcome bug reports, feature requests, and documentation fixes. Send us the problem, not the patch — a precise issue with a reproduction is more useful than a pull request.
🐛 Report a bug: Open an issue with a reproduction. The most useful thing you can send us.
💡 Propose a feature: Open an issue — the problem and who hits it, not the implementation.
🔧 Code: Work only on a maintainer-invited issue. An open issue is not an invitation to pick it up; unsolicited pull requests are closed.
📚 Documentation: Typos, broken links, and wrong commands go straight to a PR.
