Problem
Apify's MCP server already lets AI agents call Actors from the Apify Store. But if a developer builds a private or custom Actor, there is no CLI path to make it available as an MCP tool to their local AI agent.
The workflow today is:
apify push — deploy Actor to Apify
- Manually configure the MCP server to include that Actor
- Hope the AI agent discovers it
There's no single CLI command that closes this loop.
Proposal
Add a subcommand that registers a pushed Actor as an MCP-accessible tool in the user's local IDE config:
apify actors expose-as-mcp [ACTOR_ID]
Or as a flag on push:
apify push --expose-as-mcp
What it does
- Reads the Actor's
INPUT_SCHEMA.json to generate an MCP tool description
- Registers the Actor in the local MCP config (same targets as
apify mcp install)
- The Actor then appears as a callable tool in Claude, Cursor, Windsurf, etc.
Example
$ apify push --expose-as-mcp
Pushing Actor... ✓ my-custom-scraper deployed (version 0.3)
Registering as MCP tool...
Tool name: apify_my-custom-scraper
Description: (from INPUT_SCHEMA.json title/description)
✓ Added to Claude Desktop ~/.claude/settings.json
✓ Added to Cursor ~/.cursor/mcp.json
Restart your IDE. You can now say: "Run my custom scraper on https://example.com"
MCP tool shape (auto-generated from INPUT_SCHEMA.json)
{
"name": "apify_my-custom-scraper",
"description": "Scrapes product listings from e-commerce sites",
"inputSchema": { /* derived from Actor INPUT_SCHEMA.json */ }
}
Why it matters
- Turns every Apify Actor into an AI-native tool with one flag
- Direct revenue channel: each AI-triggered Actor run = compute billing
- Lowers the skill floor for AI developers who don't know Apify internals
- Differentiates Apify from generic scraping APIs — you publish once, AI agents discover automatically
Out of scope
- Auto-generating the input schema (see related issue below)
- Publishing to the public Apify Store (separate flow)
Related
Problem
Apify's MCP server already lets AI agents call Actors from the Apify Store. But if a developer builds a private or custom Actor, there is no CLI path to make it available as an MCP tool to their local AI agent.
The workflow today is:
apify push— deploy Actor to ApifyThere's no single CLI command that closes this loop.
Proposal
Add a subcommand that registers a pushed Actor as an MCP-accessible tool in the user's local IDE config:
Or as a flag on push:
What it does
INPUT_SCHEMA.jsonto generate an MCP tool descriptionapify mcp install)Example
MCP tool shape (auto-generated from INPUT_SCHEMA.json)
{ "name": "apify_my-custom-scraper", "description": "Scrapes product listings from e-commerce sites", "inputSchema": { /* derived from Actor INPUT_SCHEMA.json */ } }Why it matters
Out of scope
Related
apify/apify-mcp-serverchore(deps): update dependency @apify/json_schemas to ^0.17.0 #1351 — Actor development loop over MCPapify/apify-mcp-serverdo not add .venv to gitignore #1353 — Add build-actor toolapify mcp installfor AI IDEs