AI agents · MCP

Give your agent the live web.

One MCP server for Claude, ChatGPT, Codex, Cursor and 16 more. Your agent searches three engines, reads pages as markdown, and captures screenshots, from a browser that sites do not block.

https://api.wayfern.com/mcp
  • 20 agents, each with its own setup
  • OAuth sign-in, no token to paste
  • Every result reports its credits
A board with the app icons of 22 AI agents. A frame moves from agent to agent, and the line below names the route each one takes into Wayfern and the tool it calls.

Ask anything

Your agent calls a tool. The page comes back.

Pick a chat. Each one shows the real tool, its inputs and what it cost. The domains and answers are sample data.

Wayfern MCP Connected Sample data

Pull the brand colours from northwind.dev and give them to me as CSS variables.

Ask anything

A practical agent loop

Turn a URL into evidence the agent can check.

  1. 01

    Describe the target

    The agent picks a tool and fills in its inputs: a URL, a query, a selector, a JSON Schema.

  2. 02

    Wayfern runs it in a real browser

    An anti-detect Chromium loads the page, waits for it, removes cookie banners and retries on a residential exit when a site blocks it.

  3. 03

    The agent reads, decides, reports

    Markdown, JSON or an image URL comes back with credits_charged, so the agent knows the cost of each step.

tools/call

{
  "name": "capture_screenshot",
  "arguments": {
    "url": "https://acme.dev/pricing",
    "full_page": true,
    "block_cookie_banners": true
  }
}

Result

{
  "screenshot": {
    "url": "https://…/screenshots/8c1f….png",
    "urlExpiresAt": "2026-10-29T13:00:00.000Z",
    "contentType": "image/png"
  },
  "credits_charged": 5
}

The agent gets a link, not image bytes. It opens the image when it needs to look, and the link expires on its own.

Three ways in

The web data API for AI agents.

Each way calls the same services, on the same balance, at the same rates.

  • Agents that speak MCP

    Remote MCP server

    Add the URL and sign in on wayfern.com with OAuth. 11 tools appear. Clients that send headers can use a wf_ token.

    server

    https://api.wayfern.com/mcp
    MCP reference
  • Agents with a shell

    Agent skill

    One SKILL.md file teaches Claude Code, Codex and others to call the REST API with curl and your token.

    bash

    mkdir -p ~/.claude/skills/wayfern
    curl -fsSL https://wayfern.com/skills/wayfern/SKILL.md -o ~/.claude/skills/wayfern/SKILL.md
    Skill setup
  • Your own code

    API and browsers

    Call the Web Data API for deterministic steps, or drive a Wayfern browser with Playwright over CDP.

    bash

    curl https://api.wayfern.com/api/v1/scrape \
      -H "Authorization: Bearer $WAYFERN_API_TOKEN" \
      -H "Content-Type: application/json" \
      -d '{"url": "https://example.com"}'
    API reference

Reliable browser execution

Control the page before the agent sees it.

The same options work from the MCP tools and the REST API. The guides show each one.

  • Wait for the right state

    Wait for a CSS selector or a fixed delay. Tall pages scroll so lazy images load.

  • Remove visual noise

    Cookie banners are dismissed, ads blocked, and any selector you name is hidden.

  • Capture the needed context

    Set the viewport, emulate a phone, take the full page, one element, or slices.

  • Return machine-friendly results

    Markdown, HTML, links, JSON that matches a schema, or an image URL that expires.

FAQ

Questions about agents

Tool inputs and costs are in the MCP reference.

What is the Wayfern MCP server?

A remote Model Context Protocol server at https://api.wayfern.com/mcp. It gives an AI agent 11 tools to search the web, scrape pages, extract data, map and crawl sites, and capture screenshots, all in Wayfern's anti-detect browser.

Which clients does it support?

Any client that supports remote MCP over Streamable HTTP. This page has setup steps for 20 agents, including Claude, ChatGPT, Claude Code, Codex, Cursor, VS Code and Gemini.

Do I need an API key?

No. Clients that support MCP authorization open wayfern.com, and you sign in and click Allow. Clients that cannot do OAuth can send a wf_ token with the mcp scope in an Authorization header.

Should my agent use MCP or the API?

Use MCP when the model decides which page to read or which search to run. Use the REST API when your own code runs a fixed step, such as a nightly crawl. Both call the same services at the same rates.

Do MCP tool calls use credits?

Yes. Each tool that fetches from the live web uses credits from your account at the REST API rates, and each result reports credits_charged. list_search_engines, crawl_status and get_usage are free.

How do I remove an agent’s access?

Open Tokens in the dashboard. Connected apps lists each OAuth connection with its last use. Revoke stops the next tool call.

Why use a screenshot and not only text?

Text loses layout, colour and what is visible above the fold. A screenshot lets a vision model check that a price, a banner or a plan column is really on the page.

Does it work on pages that block bots?

Each call runs in an anti-detect Chromium with a generated fingerprint. In proxy mode auto, a request that a site blocks is retried on a residential exit. Wayfern cannot promise that every site loads.

Does it return large images to the model?

No. capture_screenshot returns a download URL, so the model context gets a short link and the client fetches the image only when it needs it.

Connect Wayfern to your agent.

Paste the server URL into your client and sign in. The tools appear at once.

https://api.wayfern.com/mcp