Web search for AI agents over MCP

The Wayfern MCP server gives an AI agent a search_web tool and a research_with_wayfern prompt. This guide shows how to connect, what the tool takes, what the prompt does and what it costs.

Connect

The endpoint is https://api.wayfern.com/mcp. It uses Streamable HTTP. Authenticate with a wf_ API token that has the mcp scope. Your account needs an active paid plan or a positive credit balance.

{
  "mcpServers": {
    "wayfern": {
      "type": "http",
      "url": "https://api.wayfern.com/mcp",
      "headers": {
        "Authorization": "Bearer ${WAYFERN_API_TOKEN}"
      }
    }
  }
}

Keep the token in your client’s secret store, not in a committed file. Tokens made before MCP support do not have the mcp scope. For client setup, see Connect an agent.

The search_web tool

search_web runs the same search as the Search API and returns the same envelope.

Input	Default	Notes
query	required	1 to 512 characters
engines	["google"]	Any of google, bing, duckduckgo
language	none	Such as en or en-US
region	none	Such as US
date	none	YYYYMMDD..YYYYMMDD
filetype	none	Such as pdf
site	none	One domain
limit	50	Results per engine, 1 to 200
start	0	Offset for the next page
filter_duplicates	true	Hide near-duplicate results
include_related	true	Return related queries in serp_features

The REST API accepts a search with only site or filetype. The tool always needs query.

When a search fails, the tool returns an error result with the text Wayfern search_web failed: and the reason. The agent can read it and try again with other inputs.

These are free: the list_search_engines tool, the wayfern://search-engines and wayfern://mcp/help resources, and the prompt.

The research_with_wayfern prompt

The prompt takes two arguments:

• topic: required, 1 to 512 characters.
• engines: optional text, such as google, bing.

It returns one user message. For the topic “headless browser pricing” and the engines “google, bing”, the message is:

Research “headless browser pricing” using the Wayfern search_web tool with these providers: google, bing. Compare multiple independent sources, follow useful related searches, and cite the result URLs.

The prompt costs nothing. The search_web calls that the agent then makes are billed.

Call the tool without a client

A raw call with the current stateless protocol:

curl -X POST https://api.wayfern.com/mcp \
  -H "Authorization: Bearer $WAYFERN_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2026-07-28" \
  -H "Mcp-Method: tools/call" \
  -H "Mcp-Name: search_web" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "search_web",
      "arguments": { "query": "headless browser pricing", "limit": 10 },
      "_meta": {
        "io.modelcontextprotocol/protocolVersion": "2026-07-28",
        "io.modelcontextprotocol/clientCapabilities": {},
        "io.modelcontextprotocol/clientInfo": { "name": "my-agent", "version": "1.0.0" }
      }
    }
  }'

Control the cost

search_web costs 2 credits for each provider page it fetches, cache hits included. Google returns up to 100 results a page. Bing and DuckDuckGo return 10.

• The default limit of 50 is 1 Google page (2 credits), but up to 5 Bing pages (up to 10 credits).
• Each extra engine fetches its own pages.
• A captcha that Wayfern solves during a search debits 1 credit more.

The search_web result does not report its cost. The Web Data tools report theirs in credits_charged. Check search spend on the Usage page of the dashboard. To cap what auto top-up can spend in a month, set a spend limit. When the limit stops a call, the tool returns an error that names the reason. See Credits and spend limits.

Tool inputs and the protocol are in the MCP reference.