subimpact data (Apify)

@subimpact/data-mcp

Public-data tools for AI agents, in one MCP server. Point it at your own Apify account and your model can search Google (organic results, People Also Ask, knowledge graph and full AI Overviews), run Google AI Mode queries, watch news mentions across Southeast Asian publishers, collect Google Maps places, Google Jobs, Trends, website contacts, page-to-markdown, Ads Transparency data, YouTube, TikTok, Eventbrite, ATS career pages, Substack, Bluesky, Lazada products, JSON-LD entity graphs, SEO audits and Pinterest — all as JSON instead of browser tabs. Every tool wraps one actor from the subimpact Apify Store: the server starts the actor with your APIFY_TOKEN, waits for it, and returns the dataset items. Nothing is scraped locally and no credentials of ours are involved.

Install

You need an Apify API token: console.apify.com → Settings → Integrations. Runs are billed to that Apify account at each actor's Store price (pay per event).

Claude Code

claude mcp add subimpact-data -e APIFY_TOKEN=... -- npx -y @subimpact/data-mcp

Claude Desktop

Add this to claude_desktop_config.json (~/Library/Application Support/Claude/ on macOS, %APPDATA%\Claude\ on Windows):

{

"mcpServers": { "subimpact-data": { "command": "npx", "args": ["-y", "@subimpact/data-mcp"], "env": { "APIFY_TOKEN": "apify_api_your_token" } } } }

Cursor

Add the same block to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{

"mcpServers": { "subimpact-data": { "command": "npx", "args": ["-y", "@subimpact/data-mcp"], "env": { "APIFY_TOKEN": "apify_api_your_token" } } } }

Want a smaller tool list? Add "--tools", "google_search,news_mentions" to args, or set SUBIMPACT_TOOLS=google_search,news_mentions in env. get_run_results is always available.

Tools

Every tool also accepts two optional control parameters:

• _maxItems (integer, default 50, max 1000) — how many dataset items come back to the model. It truncates the response only; the actor still collects (and charges for) what its input asks for.
• _maxChargeUsd (number, default 0.5) — per-run spending cap in USD, sent to Apify as the run option maxTotalChargeUsd. The run stops when it reaches the cap, so an accidental large batch cannot cost more than this.

The extra get_run_results tool (runId, _maxItems, offset) fetches the status and items of a run that was still running when its tool call returned. Long jobs return a runId instead of blocking forever; call it later to collect the results.

Cost control

Every call is a real Apify run on your account, priced per event by the actor (for example google_search at $0.003 per SERP, news_mentions at $0.001 per article). There is no subscription and this server never charges you: Apify bills your account directly.

• Start with the default _maxChargeUsd of $0.50; it is passed to Apify and enforced by the platform.
• Use _maxItems to keep model context (and token spend on the model side) small.
• Narrow the actor input instead of raising the cap when you need fewer, more targeted results.
• Each response includes itemCount (total items in the dataset) and returned (items sent to the model), so you can see when a result was truncated.

A single tool call waits up to SUBIMPACT_TIMEOUT_SECS (default 240 seconds). If the run is still going, you get its runId and can collect the results with get_run_results.

Configuration

CLI flags: --tools a,b,c, --help, --version.

What it is not

• Not a local scraper. It does not fetch pages or parse HTML itself; it runs Apify actors, which do the scraping on Apify's infrastructure.
• Not a credential holder of ours. The only token it uses is the APIFY_TOKEN you set. The server ships with no secrets and calls no private backend.
• Not free. Every tool call is an Apify run billed to your Apify account at the actor's Store price. The _maxChargeUsd cap limits a single run, not your monthly Apify spend.
• Not an official Apify product, and not affiliated with Google, TikTok, Meta, Lazada or the other platforms whose public data the actors read.
• Not a guarantee of data. Public sites change; review each actor's Store page for current coverage, pricing and limitations.

Development

npm install

npm run build # tsc + copy src/tools.json into dist/ npm test # offline tests: mocked fetch + a stdio round trip npm run typecheck