A scraping API your agent can call, and pay for, per result

What does a pay-per-result scraping API cost an AI agent?

Every Datapika run bills per delivered result and nothing else. There is no monthly plan, no seat and no minimum: a free Apify account is enough to start, and that account is charged for the rows an actor returns. The rate is set per actor. As of August 2026 the Apify Store lists the hiring signals scanner at $0.0005 per record, the Google Flights actor at $0.0015 per itinerary, the Reddit actor at $0.002 per post, the Google Maps actor at $0.003 per place and the multi-board job actor at $0.005 per job. Trend reports are priced per report instead, at $0.35 for the standard tier. Because the unit is a row, an agent can compute the cost of a call before making it and skip calls that exceed its allowance.

• A 100-job sweep on the job board actor costs $0.50; 1,000 hiring-signal company checks also cost $0.50
• 1,000 Google Flights itineraries cost $1.50; a 1,000-place Google Maps run is $3.00 at a flat $0.003 per place
• maxTotalChargeUsd on any REST run stops the platform at that ceiling instead of overspending
• The TikTok actor returns an unbilled error row for a failed URL, and the Airbnb actor charges nothing for empty or failed runs, so retries do not compound cost

How does an AI agent connect to Datapika over MCP?

The actors are exposed through the Apify MCP server over streamable HTTP with OAuth. One URL preloads every Datapika tool by listing each slug after fetch-actor-details, for example https://mcp.apify.com/?tools=fetch-actor-details,openclawai/job-board-scraper,openclawai/reddit-scraper and so on through the catalog. Each actor also has a minimal single-tool URL of the form https://mcp.apify.com/?tools=fetch-actor-details,openclawai/<slug>, which is the right choice when an agent needs only one source. The paired fetch-actor-details tool is what makes the setup agent-safe. Before running anything, the model can read the actor's input schema and current price and decide whether the call fits the budget it was given. On first use the client opens Apify OAuth in the browser; a free account completes it, and every run after that bills per result to the same account.

• Claude Code: claude mcp add --transport http datapika "<mcp url>"
• Cursor: the Add to Cursor button on datapika.com/mcp, or the URL in .cursor/mcp.json
• VS Code: .vscode/mcp.json with type http; Codex CLI: codex mcp add datapika --url "<mcp url>"
• Example prompt: find remote senior data engineer roles posted in the last 24 hours, dedupe across boards, return salary ranges as a table

When should an agent use the REST API instead of MCP?

Use REST when the agent runs unattended, on a schedule, or inside a pipeline that has no MCP client. A single POST to https://api.apify.com/v2/acts/openclawai~<slug>/run-sync-get-dataset-items?format=json with a Bearer token returns dataset rows directly as JSON. The synchronous endpoint returns 408 after 300 seconds, so large sweeps should start a run, poll its status, then read the dataset when it finishes. Official Node and Python clients wrap both patterns, and the Apify CLI covers shell scripts. Each actor publishes an OpenAPI spec for its default build at https://apify.com/openclawai/<slug>/api/openapi with no auth required, so codegen or a tool-calling framework can build a typed client from it. Schedules and webhooks are platform features: run any actor every morning and push new rows to your stack.

• Sync: run-sync-get-dataset-items returns rows in one response for runs that finish inside 300 seconds
• Async: actor.call() then dataset.listItems() in the Node or Python client, required past the 300-second limit
• Budget: append maxTotalChargeUsd=<usd> to the run URL to cap spend per call
• Auth: an Apify API token in an Authorization: Bearer header, issued from a free account

How does an agent discover Datapika's tools without a human?

Three plain-text surfaces describe the catalog in a form a model can read directly. datapika.com/llms.txt follows the llms.txt convention: a one-paragraph description of the studio, every actor with its price and run URL, the MCP config URL, the guides, and a facts block. datapika.com/llms-full.txt carries every guide and actor description in one file for retrieval. datapika.com/AGENTS.md uses the installation, configuration and usage layout that coding agents already expect from repository AGENTS.md files: the MCP add command, the Bearer auth rule, the budget parameter, the 300-second timeout, and one line per actor with slug, price, docs link, input schema link and OpenAPI link. All three render from the same catalog data as the pages, so a price never drifts between what a human reads and what an agent reads.

• llms.txt: catalog map with per-actor price and run URL, plus the one-line MCP install
• AGENTS.md: installation, configuration and usage sections, including timeouts and budget caps
• Input schema at https://apify.com/openclawai/<slug>/input-schema; OpenAPI at https://apify.com/openclawai/<slug>/api/openapi
• Pages, sitemap, llms.txt and AGENTS.md share one data source, so a price change updates everywhere at once

Questions, answered

Keep reading