FourA MCP server

Start with foura_auto: one URL in, typed content out. It chooses the appropriate fetch method and handles common blocks for you. Use foura_single, foura_proxy, or foura_browser when you want explicit control over the request path.

Start simple	Reach difficult pages	Build predictable agent flows
One default tool for most URLs	HTTP, proxy rotation, and full-browser rendering	Typed outputs, stable error codes, and six ready-made prompts

The included workflows cover product extraction, article cleanup, price monitoring, endpoint checks, and bulk URL fetching.

Use it from Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, or any MCP client. Run the npm package locally or connect to the hosted Streamable HTTP endpoint with the same FourA API key.

Get an API key | Read the MCP docs | See the package on npm

One-click install:

Both buttons pre-fill the config with a YOUR_FOURA_API_KEY placeholder - replace it with your key. Or by hand: claude mcp add foura -- npx -y @fouradata/mcp (set FOURA_API_KEY in env first). Full per-client setup below.

Quick Start - local stdio (recommended for Claude Desktop)

Create or reveal an API key at foura.ai/dashboard/#api-keys (format pk_live_...). Then drop this into your MCP client's config:

{
  "mcpServers": {
    "foura": {
      "command": "npx",
      "args": ["-y", "@fouradata/mcp"],
      "env": {
        "FOURA_API_KEY": "pk_live_..."
      }
    }
  }
}

Claude Desktop gotcha: fully quit Claude Desktop (Cmd+Q on macOS) before editing the config file. If the app is still running, it will overwrite your edits with its in-memory config on exit.

The npx command downloads the package on first launch and runs it as a subprocess of your MCP client. No global install needed. The same JSON works across MCP clients; only the config-file location changes:

Client	Where the config lives
Claude Desktop (macOS)	~/Library/Application Support/Claude/claude_desktop_config.json
Claude Desktop (Windows)	%APPDATA%\Claude\claude_desktop_config.json
Claude Code	claude mcp add foura -- npx -y @fouradata/mcp (set FOURA_API_KEY in env first)
Cursor	~/.cursor/mcp.json
Windsurf	~/.codeium/windsurf/mcp_config.json
VS Code (MCP extension)	.vscode/mcp.json in your workspace

Restart the client and foura_auto, foura_single, foura_proxy, foura_browser show up in your tool list, plus six prompts under /prompts.

Quick Start - hosted (Streamable HTTP)

If your client supports the Streamable HTTP transport, point it at the hosted endpoint instead of running a local subprocess:

{
  "mcpServers": {
    "foura": {
      "url": "https://mcp.foura.ai/mcp",
      "headers": {
        "Authorization": "Bearer pk_live_..."
      }
    }
  }
}

For Claude Desktop, use the stdio config above. You can also bridge the hosted endpoint through mcp-remote:

{
  "mcpServers": {
    "foura": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.foura.ai/mcp", "--header", "Authorization: Bearer pk_live_..."]
    }
  }
}

The Tools

foura_auto is the default - give it a URL and it returns the content, choosing the fetch method for you. The other three tools give you direct control over HTTP, proxy rotation, and browser rendering.

All four are marked readOnlyHint: true and openWorldHint: true per the MCP specification. Clients can use those annotations when deciding how to present or approve tool calls.

Every response carries human-readable text (content) plus a typed structuredContent JSON object validated against the tool's outputSchema. Your agent gets a predictable response contract instead of parsing prose.

foura_auto - smart fetch (the default)

Give it a URL and get the content back. Use this first when you don't want to choose between HTTP, proxy rotation, and browser rendering. It handles common bot challenges and keeps the client surface small, with no proxy-attempt tuning required.

{
  "url": "https://example.com",
  // optional: a substring the real page must contain, so auto can tell a
  // real page from a challenge page on protected targets
  "validate": { "data": { "accept": ["Example Domain"] } }
}

The client surface is intentionally minimal: url (required), plus optional method, headers, data, validate, returnSession (default true), forceProxy (default true), timeout_ms (5000-180000, default 120000), ignoreProxies, exitClass (standard or premium, as on foura_proxy).

structuredContent shape: {status, headers, data, meta, session}. meta is always present - {rung, solved, attempts, credits} - so your agent can see how the request completed and how many credits it used. session ({proxy, cookies, userAgent}) is returned by default for follow-up requests (pass session.proxy into the proxy field). Send returnSession: false to omit it. There is no total_time field on auto.

exitClass: "premium" lets auto escalate to a premium exit once the standard pool cannot reach the page, and the result's exitClass says which class delivered. The standard pool goes first; once the premium traffic in your plan is spent the call goes on in the standard pool and reports standard. Neither class is an error. On a plan without premium exits the call is refused with code: "plan_limit_premium". exitClass: "standard" keeps auto off premium exits.

foura_single - fast HTTP

One HTTP request, response back. Use it for static pages, JSON APIs, and server-rendered HTML. unblocker defaults to true, so the request already carries a full browser header set; set it to false for a plain HTTP request.

{
  "method": "GET",
  "url": "https://example.com"
}

Supports custom headers, a body, per-stage timeouts, redirect controls, JSON auto-parse, a binary-buffer mode, and built-in response validation (validate.status.accept, validate.data.fail, and so on). If foura_single comes back blocked - status 403/429, a captcha page, challenge response headers, or a known challenge title - try a different browser profile (below), then escalate to foura_proxy. Start with maxTries: 5; protected targets may need 25-30. If the page also needs JavaScript to render, pass foura_proxy's returned proxy ID to foura_browser.proxy.

structuredContent shape: {status, headers, data, total_time, ...}. When the target ran a bot check, the response also carries defense; defense.solved: false means the body may be a challenge page rather than the content you asked for.

Choosing which browser you present

By default a request presents the latest Google Chrome. Some targets accept one browser and refuse another, so foura_single and foura_proxy let you pick:

{
  "method": "GET",
  "url": "https://example.com",
  "browser": "Firefox",
  "os": "Windows"
}

• browser - Chrome, Edge, Safari, Firefox, or Tor.
• os - Windows, macOS, Android, or iOS. A family name accepts any of its versions; the exact labels in the catalogue narrow further.
• version - any version the catalogue lists for that browser. The newest match wins when several fit.
• profile - an exact id from the catalogue, when you already have one.

The full catalogue is public at api.foura.ai/api/profiles, no key required, and its default field names the profile a request gets when you ask for nothing. Selection needs unblocker on, which is the default. A combination that does not exist returns an error listing what is available, so a request is never quietly sent as a browser you did not choose.

foura_proxy - rotating proxies with retry

Same target shape as foura_single, but routed through rotating proxies with automatic retry on failure. Use it when a direct request is blocked or when the target requires a specific exit country.

{
  "maxTries": 5,
  "exitCountries": ["CZ", "GB"],
  "request": {
    "method": "GET",
    "url": "https://example.com/pricing",
    "browser": "Chrome",
    "os": "Windows"
  }
}

structuredContent adds proxy (the encoded ID of the proxy that succeeded) and total (outer timing including selection and retries). When the rotation had to move to another browser to get an answer, profile names the catalogue profile it settled on; pass it as profile to present the same browser on the next call. exitCountries is optional: values are trimmed, uppercased, and deduplicated; proxies with unknown exits are excluded; the request never falls back to an unrequested country. Selection uses the latest available country metadata, normally updated within ten minutes, and a scoped success returns that value as exitCountry. If no eligible proxy matches, the result uses code: "no_eligible_proxy".

The inner request takes the same browser-profile fields as foura_single: browser, os, version, or an exact profile.

exitClass - for a target the standard pool cannot reach. Set exitClass: "premium" and the request may escalate to a premium exit instead of only rotating within the standard pool. It is an allowance, not an instruction: the standard pool still races for the answer and usually wins, and a request it answers before any premium exit is tried costs no premium traffic. Once a premium exit has been tried, the traffic it carried counts toward your premium allowance even if that attempt failed; an attempt still running when another exit answered is stopped and not counted. The response reports exitClass back, premium or standard, so you can see per request which one served you. standard is also the answer once the premium traffic in your plan is spent, and it is a normal result, not an error. exitClass: "standard" forbids escalation outright. On a plan without premium exits the call is refused with code: "plan_limit_premium".

attemptReport - why a rotation ran out of tries. Every failed foura_proxy result carries it beside the error. summary is one sentence; the counts underneath separate causes with opposite fixes: noResponse (the exit never answered), defense (a bot check was recognised, named in vendors), contentRejected (HTTP 200, no bot check, rejected only by your own validate.data), statusRejected, and other. profilesTried lists the browsers the task sent, in first-use order, with default meaning the request went out exactly as written. A high contentRejected means the pages arrived and your own rule threw them away: fetch one with foura_single and no validate, look at it, and rewrite the rule.

For difficult WAF challenges, use maxTries: 25-30. For a country allowlist, set exitCountries instead of increasing attempts. If the target needs JavaScript rendering, pass the returned proxy ID to foura_browser.proxy.

foura_browser - full browser session

A real browser session. JavaScript runs, the DOM finishes rendering, cookies come back with the response. Use it when the page is a single-page app, when content lazy-loads after first paint, or when there's an anti-bot challenge that needs a real browser to clear.

{
  "url": "https://example.com/spa",
  "timeout_ms": 15000,
  "checkText": "data-table"
}

Use this when an HTTP response isn't enough. checkText validates the rendered HTML once navigation completes; it doesn't wait or poll. If the substring is missing, the call fails with an error envelope. This catches pages that return 200 without the content you need. unblocker defaults to true and handles supported anti-bot or captcha challenges along the way. Set it to false to return the page exactly as it loads, challenge included.

structuredContent shape is intentionally different from single/proxy: {status, headers (object, not array), body (not data), cookies (full browser cookie shape), userAgent}.

Built-in Prompts

Six workflow templates surfaced under /prompts in your MCP client. They orchestrate one or more tools without you spelling out the steps.

Prompt	Arguments	What it does
smart_fetch	url, must_contain?, extract?	Pick the method, handle bot protection, and return or extract content
scrape_product_page	url	Extract title, price, image, stock, and SKU as JSON in a browser session
extract_article	url	Fetch an article, retry through a proxy if needed, and remove page chrome
monitor_pricing	url, target_price?	Fetch the current price and compare it with a target
check_endpoint_health	url, expected_text?	Validate reachability, status, content, and timing
bulk_fetch_urls	urls (comma-separated)	Fetch URLs in parallel, retry blocked ones, and return metadata

Each prompt arrives as a templated user message your LLM executes with the right tools. The full prompt text enters the context only when you invoke it.

Full recipe text + manual fallback prompts: foura.ai/docs/mcp/recipes. For the full error code list, see foura.ai/docs/mcp/errors.

Authentication

Use FOURA_API_KEY in stdio mode or an Authorization: Bearer pk_live_... header with the hosted endpoint. One key authenticates all four tools.

Keys are managed in the dashboard, where you can create, reveal, rotate, or deactivate them. See foura.ai/docs/getting-started/authentication for the full key-management walkthrough.

Error envelope - typed contract for agent retries

Every error (isError: true) carries a structuredContent envelope with at minimum these three fields:

{
  "service": "auto" | "single" | "proxy" | "browser",
  "code": "ssrf_blocked" | "auth_failed" | "rate_limited" | ...,
  "error": "Human-readable message"
}

Where the upstream returned a status, you also get status (HTTP code) and on rate-limit / capacity errors the FourA API envelope adds retryAfter, current.{concurrency, rpm}, limits.{maxConcurrency, maxRpm}.

code	When	Retry safe?
ssrf_blocked	Target resolves to a private or reserved address	No - change the URL
upstream_non_json	Upstream returned malformed body	Maybe - investigate
bad_request (400)	Input shape rejected by FourA	No - fix arguments
auth_failed (401)	Key missing, invalid, or deactivated	No - fix the key
forbidden (403)	Authenticated but not allowed	No
not_found (404)	Target / endpoint doesn't exist	No
rate_limited (429)	RPM cap hit	Yes - wait retryAfter
at_capacity (503)	Concurrency cap hit	Yes - wait retryAfter
service_disabled (503)	Maintenance window	Yes - wait retryAfter
service_unavailable (503)	Generic 503	Yes - short backoff
upstream_error (>=500)	Upstream 5xx	Yes - exponential backoff
upstream_client_error (4xx)	Other 4xx	Usually no
no_eligible_proxy	No proxy matches the requested exitCountries	No - change the country scope
plan_limit_* (403/429)	Your own FourA plan refused the call, not the target	Only retryAfter clears it - see below

A plan_limit_ code names which of your plan's limits refused the call: credits, bandwidth, rate, concurrency, browser_daily, premium, or feature. It is not the target blocking you, so retrying the same work through another tool spends the rest of the allowance for nothing. Wait out retryAfter where there is one, or change the plan.

LLM agents can read code directly for retry logic without parsing prose. Spec reference: foura.ai/docs/api/errors.

What every result carries

Beside the tool's own fields, each result reports what the call cost and how to trace it:

• credits - credits this call spent, on failures as well, because the work was done either way;
• request_id - FourA's id for the call. Quote it in a support request;
• exitClass - premium when a premium exit served the call. On foura_single and foura_browser that happens when proxy replays a premium exit that foura_proxy or foura_auto found; on foura_proxy and foura_auto it is reported whenever the request named an exitClass.

Combining the tools - reuse the same exit

The lower-level tools compose. foura_proxy returns the base36 ID of the exit it used. Pass that ID into foura_single.proxy or foura_browser.proxy to reuse the same exit for the next request.

// 1. Find a working exit. Difficult protected targets may need maxTries:25-30.
const r = await foura_proxy({
  maxTries: 30,
  request: { method: "GET", url: "https://probe.example.com" }
});
// Returns { status: 200, proxy: "4DZ3VE", ... }

// 2. Reuse it for follow-up HTTP
await foura_single({ method: "GET", url: "https://target/api", proxy: r.proxy });

// 3. Or render JavaScript through the same exit.
await foura_browser({ url: "https://target/spa", proxy: r.proxy });

For a protected JavaScript page, find a working route with foura_proxy before calling foura_browser. Starting in the browser can return the challenge page before the real content loads.

To choose a different proxy on the next foura_proxy call, pass the previous ID as ignoreProxies: ["4DZ3VE"]. The proxy field on foura_single and foura_browser also accepts raw URLs (http://host:port, socks5://...) if you have your own list.

Large responses - offload_large (default: inline)

By default (since v0.2.0), full response bodies are returned inline in structuredContent regardless of size. This works in every MCP client.

If your client supports MCP resources/read (and you want to save tokens on big pages), pass offload_large: true per tool call. Responses of 50 KB or more are returned as a resource_link, and your client fetches the body only when it needs it. The resource remains available for one hour.

{
  "method": "GET",
  "url": "https://en.wikipedia.org/wiki/Web_scraping",
  "offload_large": true   // opt in for token savings
}

Support for MCP resource links varies by client. Leave offload_large at its default false if your client can't read resource_link blocks.

Only the API key that created an offloaded resource can read it back.

Limits and responsible use

• Private targets are refused. Requests to private or reserved addresses are blocked at the MCP layer. Only public-internet hosts are forwarded.
• Rate limits are enforced by the FourA API per service. Concurrency + RPM. Details at foura.ai/docs/api/rate-limits.

Use FourA only for public content you are authorized to access and in accordance with the target site's terms and applicable law.

Run it yourself

Run the stdio server directly from npm with npx -y @fouradata/mcp, or build the included Dockerfile for the Streamable HTTP transport.

Configurable environment:

Variable	Default	Purpose
PORT	3076	HTTP listen port
FOURA_API_BASE	https://api.foura.ai/api	Upstream FourA REST base URL
FOURA_MCP_PAYLOADS_DIR	/data/payloads	Where responses of 50 KB or more are cached on disk

See DEVELOPMENT.md for the local build and test workflow.

License

MIT. See LICENSE.

Links

• FourA (web scraping API): https://foura.ai
• MCP server page: https://foura.ai/mcp
• Source (GitHub): https://github.com/fouradata/mcp
• npm package: https://www.npmjs.com/package/@fouradata/mcp
• API documentation: https://foura.ai/docs
• MCP server reference: https://foura.ai/docs/mcp/server
• MCP error codes: https://foura.ai/docs/mcp/errors
• MCP recipes: https://foura.ai/docs/mcp/recipes
• REST API errors: https://foura.ai/docs/api/errors
• MCP specification: https://modelcontextprotocol.io
• Get a key: https://foura.ai/dashboard/#api-keys