# Flowise (/docs/integrations/flowise) Pull social media data into Flowise without writing a server. You can call SocialCrawl from a Flowise agent with a Custom Tool. The tool runs a short JavaScript function that calls the API and returns the answer to the model. Setup takes three steps: store the key as a variable, write the tool, attach it to the agent. ## Before you start Create an API key in the dashboard under API Keys. In Flowise, add it as a variable named `socialcrawlKey`, and read it in the tool as `$vars.socialcrawlKey`. Never write it into the function. Prove the key with a call that costs nothing: ```bash title="Check the key (0 credits)" curl "https://www.socialcrawl.dev/v1/credits/balance" \ -H "x-api-key: sc_your_key_here" ``` A reply with `"success": true` and your balance in `data.balance` means the key works. ## Connect Create a Custom Tool with one input property, `handle`. In the function, an input property is available as a variable with a `$` prefix, so `handle` becomes `$handle`. The function must return a string. ```javascript title="Custom Tool function" const url = "https://www.socialcrawl.dev/v1/tiktok/profile?handle=" + encodeURIComponent($handle); const res = await fetch(url, { headers: { "x-api-key": $vars.socialcrawlKey }, }); const body = await res.json(); if (!body.success) { return `SocialCrawl refused the call: ${body.error.type}. ${body.error.message}`; } return JSON.stringify(body.data); ``` Change the path to call any other endpoint, and add input properties for its parameters. ## Read the answer Every response has the same shape. Branch on `success`. A list puts its rows in `data.items`, and `pagination.next_cursor` is the token for the next page. ```json title="Response (trimmed)" { "success": true, "data": { "author": { "username": "charlidamelio" } }, "credits_used": 1, "credits_remaining": 99 } ``` To read the next page, give the tool a `cursor` input and send it as the `cursor` query parameter. Stop when `pagination.has_more` is `false`. See [Pagination](/docs/pagination.md). ## When it stops A Custom Tool hands its returned text to the model, so the function above returns the refusal as text and the model tells the user. The tool never retries. When credits run out, the answer is a 402 with `error.type` set to `INSUFFICIENT_CREDITS`. Return `error.message`, and add `error.details.checkout_url`, the link that opens checkout. When the link is absent, `error.details.top_up_url` opens the billing page. Never retry a 402, because it fails until credits are added. Retry a 429 after the seconds in the `Retry-After` header, and a 502, 503 or 504 after a short backoff. See [Errors](/docs/errors.md). ## What a run costs It depends on the endpoint, and some endpoints are priced by what they return. [Endpoint pricing](/docs/endpoint-pricing.md) lists each one, and [Credits](/docs/credits.md) explains cache hits and refunds. An agent that loops can call a tool many times in one conversation. `GET /v1/utility/estimate` quotes a call for 0 credits. - [Errors](/docs/errors.md): Every error code, with the fix and whether to retry. - [Skills & MCP](/docs/skills-and-mcp.md): Give an agent every endpoint as a tool. - [Endpoint pricing](/docs/endpoint-pricing.md): What each endpoint costs before an agent calls it.