MCP Server
Plug NeuroAPI directly into any MCP-compatible AI agent — Claude Desktop, Cursor, Continue, ChatGPT, Windsurf, VS Code Copilot, or your own — and let the agent scrape, search, map, crawl, and extract the web on demand.
Endpoint
Streamable HTTP transport. Your API key goes directly in the URL path — no auth header needed. This makes MCP clients that don't support custom headers (ChatGPT connectors, Cursor, Zed) work out of the box.
Content-Type: application/json
Streamable HTTP (stateless) · CORS enabled for all origins.The API key is embedded in the URL path. Only paste the full URL into trusted MCP clients and never commit it to a public repo. You can rotate a key anytime from the dashboard.
Available tools
All tools are namespaced with the neuroapi_ prefix to avoid collisions with other MCP servers your agent has connected (e.g. neuroapi_scrape vs another server's scrape).
Fetch a single URL as clean Markdown. 1 credit.
Web search with optional result scraping. ~2 credits.
Discover every URL on a website. 1 credit.
Start a multi-page crawl job. Reserves credits up-front.
Poll a crawl job until done. Free.
Cancel a running crawl; refunds unused credits.
LLM-powered structured JSON from a URL. 2 credits.
Capture a PNG of a URL. 1 credit.
AI summary of a page in one call. 1 credit.
Ask a natural-language question about a page. 5 credits.
Extract relevant source text from a page matching a query. 5 credits.
Scrape many URLs in one job. 1 credit per URL.
Poll a batch scrape job. Free.
Click / type / scroll on a prior scraped page (uses sessionId). 2 credits.
Release an interact session early. Free.
Parse a PDF/DOCX/XLSX/HTML (≤50 MB) into markdown — pass a `url` or `fileBase64`. 1 + 1/page.
Run a full autonomous browser agent against a URL with a natural-language task. 5 credits.
Search academic papers via the Research Index. 1 credit.
Fetch metadata + abstract for a single paper. 1 credit.
Find GitHub repos related to a research query. 1 credit.
Return current credits balance. Free.
Interacting with a scraped page
Call neuroapi_scrape first and keep the returned sessionId. Then call neuroapi_interact with either a prompt or code. If your agent sends both, blank code is ignored and prompt wins. Timeouts are accepted in milliseconds. Sessions are scoped to the API key that created them — another user's sessionId will return not-found.
{ "sessionId": "UPSTREAM_KEY_ID.SCRAPE_ID", "prompt": "Click the Browse Gallery link, wait for navigation, then return the updated page content.", "timeout": 30000}If the session has expired, call neuroapi_scrape again to get a fresh sessionId.
Claude Desktop
Add to claude_desktop_config.json (Settings → Developer → Edit Config) and restart Claude.
{ "mcpServers": { "neuroapi": { "url": "https://neuroapi.me/api/nk_live_YOUR_KEY/mcp", "transport": "http" } }}Cursor / Continue
Add to your MCP config (Cursor → Settings → MCP, or ~/.continue/config.json).
{ "mcpServers": { "neuroapi": { "url": "https://neuroapi.me/api/nk_live_YOUR_KEY/mcp" } }}Raw JSON-RPC
Talk to the server directly to debug or integrate with custom agents.
curl -X POST https://neuroapi.me/api/nk_live_YOUR_KEY/mcp \ -H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {} }'