MCP FOR AI AGENTS

Connect AI to search.
Ground every answer.

Bring Google web search into your AI client, with source URLs for every result.

01
Create an API key
Create a key in the console and use your existing credits.
02
Add your client configuration
Copy the configuration and provide your API key.
03
Start searching
Enable google_search to find web pages and source URLs.

Create API key MCP documentation

Connect your AI client

Streamable HTTP

MCP ENDPOINT

https://serpcheap.com/mcp

Cursor · mcp.json

{
  "mcpServers": {
    "serpcheap": {
      "headers": {
        "Authorization": "Bearer ${env:SERP_API_KEY}"
      },
      "url": "https://serpcheap.com/mcp"
    }
  }
}

Claude Code · Terminal

claude mcp add --transport http serpcheap https://serpcheap.com/mcp --header "Authorization: Bearer ${SERP_API_KEY}"

VS Code · mcp.json

{
  "inputs": [
    {
      "description": "SERP API Key",
      "id": "serp-key",
      "password": true,
      "type": "promptString"
    }
  ],
  "servers": {
    "serpcheap": {
      "headers": {
        "Authorization": "Bearer ${input:serp-key}"
      },
      "type": "http",
      "url": "https://serpcheap.com/mcp"
    }
  }
}

Codex · Terminal

codex mcp add serpcheap --url https://serpcheap.com/mcp --bearer-token-env-var SERP_API_KEY

Configure the credential as shown for your client. Never publish your API key.

One API key, shared credits

1 credit per successful search

Failures & tool discovery are free

Requires Streamable HTTP with custom authentication headers. OAuth-only clients are not supported.

MCP reference Understand credit pricing

ONE TOOL. STRUCTURED SEARCH.

A search tool your agent can use.

MCP lets a compatible AI client discover and call google_search. Your client sends the query; SerpCheap returns structured search data; the model uses it to organize an answer with sources.

google_search

Google web · First page

qRequired: Search query, up to 2,000 UTF-8 bytes.
glDefault: us: Country setting, for example us or de.
hlDefault: en: Search language, for example en or de.

Read the MCP contract

Static example · Tool arguments, not a full MCP protocol request

google_search · JSON

{
  "q": "Go programming documentation",
  "gl": "us",
  "hl": "en"
}

What the result gives your client

result.organic · Titles, links, snippets and original positions.
result.searchParameters · q / gl / hl
requestId / httpStatus · Request tracing and status.
result.credits · One credit per successful request.

Related questions, related searches, knowledge graphs and other modules appear only when present. There is no fixed result count.

SEARCH INSIDE YOUR WORKFLOW

Start with a question worth searching.

Explore use cases

Copy a prompt into a configured client. These examples do not run on this page; actual successful searches use your credits.

Research with sources

Find relevant pages, then let your model compare and summarize them.

Use google_search to research practical uses of retrieval-augmented generation. Search with gl=us and hl=en, and cite the source URLs in your summary.

Find developer documentation

Discover official guides and examples for your next implementation.

Use google_search to find the official Go context documentation with gl=us and hl=en. List relevant pages and their source links.

Compare regional results

Run separate searches with explicit country and language settings.

Use google_search to search for electric bikes with gl=us, hl=en, then gl=de, hl=de. Compare the returned sources. Each successful search uses one credit.

Your AI client

Interprets your request and chooses when to call the tool.

SerpCheap search

Authenticates the key and runs the search with q, gl and hl.

A sourced answer

Returns data to the model; the client organizes and checks the answer.

ONE ACCOUNT. ONE BALANCE.

Use MCP with the credits you already have.

No separate MCP subscription. REST, MCP and Playground use the same API keys, balance and request ledger.

Successful search, including valid empty results: 1 credit
Tool discovery and final search failures: 0 credits
Purchased credits: Never expire

See pricing and credit rules

CONNECTION HELP

Connected? Here is what to check next.

Keep the request ID when reporting a failed search. The documentation explains authentication, status codes and request tracing.

Error reference

Why is google_search not visible?

Check the /mcp endpoint, client configuration and Authorization: Bearer header. Replace placeholder credentials, enable or reconnect the server, then refresh the client’s tool list.

Can I use an OAuth-only connector?

Not currently. This service uses API key authentication. Choose a client that accepts Streamable HTTP with custom headers, or integrate the REST API from your application.

Why does a connected tool still fail?

Tool discovery does not validate search availability or balance. Check the MCP tool result’s isError and httpStatus, the API key, remaining credits and rate limits; connection success alone is not search success.

Will a model calling the tool several times use several credits?

Each successful search costs one credit, including repeated queries. A task can make several searches. Check the request log and credit ledger to understand its usage.

What should I do after a timeout or a 429?

After an uncertain timeout, check your request log and credit ledger before submitting again. Each new successful request is charged. For rate limits, respect Retry-After when provided and reduce concurrency; avoid unlimited automatic retries.

Do failed searches consume credits?

Final failures cost zero. Any reserved credit is released. Each successful request costs one credit, including valid empty results.

Give your AI client a search tool.

Create a key, copy the configuration above, and try a query with your own credits.

Create API key MCP reference