MCP server

SerpResult is also a remote MCP server, so an AI client can run a live search and read back the same JSON the REST API returns. There is one endpoint, POST https://api.serpresult.com/mcp, and one tool, serp_search. The transport is Streamable HTTP and stateless: nothing to install, nothing to run locally, no package to keep up to date. Point a client at the URL and it discovers everything else.

https://api.serpresult.com/mcp

The dashboard page at https://app.serpresult.com/dashboard/developers/mcp has the same setup snippets with your key filled in, and the Connected apps list.

Authenticate

Two ways in, and a reader picks one. Sign in for a personal client; use a key for a committed config or a server.

	Sign in	API key
How	The client discovers OAuth from the endpoint and opens the browser. You approve once, scope search.	Authorization: Bearer sk_live_… or sk_test_…, the same key you use for HTTP. See Authentication.
Fits	claude.ai, Claude Desktop, ChatGPT, Claude Code on your own machine.	A .mcp.json checked into a repo, a shared Cursor or Codex config, a server-side agent.
Lifetime	Access tokens last 1 hour and refresh silently for 30 days.	Until the key is rotated or revoked.
Revoke	Connected apps in the dashboard.	Revoke the key under Developers → API Keys.

A connection made by signing in draws credits straight from your account and never creates a key, so nothing shows up under API Keys for it.

Set up your client

Claude.ai & Desktop

1. Open Customize → Connectors and press Add custom connector.

2. Paste https://api.serpresult.com/mcp as the URL and press Connect.

3. Sign in to SerpResult in the window that opens and approve the connection.

An API key works here too, through the request-header option, but that option is an organisation-limited beta. Its value must include the literal Bearer prefix: Bearer sk_live_…, not the bare key.

ChatGPT

1. Turn on Settings → Security and login → Developer mode.

2. Open Settings → Apps & Connectors and press Create.

3. For Icon, upload the SerpResult icon. ChatGPT shows a generic icon when this is left empty.

4. Paste https://api.serpresult.com/mcp as the URL, choose OAuth for Authentication, and sign in to SerpResult when asked.

Available on Pro, Plus, Business, Enterprise and Edu plans.

Claude Code

Add the server, then run /mcp in a session to sign in:

claude mcp add --transport http serp-api https://api.serpresult.com/mcp

To use a key instead of signing in:

claude mcp add --transport http serp-api https://api.serpresult.com/mcp \
  --header "Authorization: Bearer sk_live_your_key"

For a config committed with the project, put this in .mcp.json at the repo root and set SERP_API_KEY in the environment:

{
  "mcpServers": {
    "serp-api": {
      "type": "http",
      "url": "https://api.serpresult.com/mcp",
      "headers": { "Authorization": "Bearer ${SERP_API_KEY}" }
    }
  }
}

Cursor

Add the server to .cursor/mcp.json in the project, or to ~/.cursor/mcp.json for every project:

{
  "mcpServers": {
    "serp-api": {
      "type": "http",
      "url": "https://api.serpresult.com/mcp",
      "headers": { "Authorization": "Bearer ${SERP_API_KEY}" }
    }
  }
}

Codex CLI

Add the server to the Codex MCP config in the same shape:

{
  "mcpServers": {
    "serp-api": {
      "type": "http",
      "url": "https://api.serpresult.com/mcp",
      "headers": { "Authorization": "Bearer ${SERP_API_KEY}" }
    }
  }
}

The serp_search tool

The tool takes the GET /v1/query parameters as its arguments and answers that endpoint’s JSON response verbatim: once as structuredContent, and once more as pretty-printed text so a client that only reads text still gets the whole thing. The response shape is documented on the endpoint page; the countries geo accepts are on Targets & geo.

Argument	Type	Default	Notes
query	string	required	The search terms, 1 to 400 characters.
target	string	google_search	Any target on Targets & geo.
geo	string	us	ISO 3166-1 alpha-2, one of the 193 supported countries.
locale	string	from geo	Overrides the country’s default locale.
page	integer	1	1 to 5 where the target supports paging; the others serve page 1 only.
num	integer	10	1 to 20. duckduckgo_search and yahoo_search cap at 10.
safe	string	off	off, moderate, strict.

The per-target table on Targets & geo says which target takes which page, num and safe values; a combination the target cannot serve is refused, not silently ignored.

Limits and cost

Every serp_search call costs 1 credit, exactly like a GET /v1/query. The tool description tells the model so, which is why a well-behaved client searches once per question rather than re-running variations.

Calls made with an API key share that key’s per-minute bucket with its HTTP traffic: an MCP call and a GET /v1/query on the same key draw on the same counter, at the key’s own limit. An app connected by signing in gets 60 calls a minute per account. Rate limits has the full table.

Troubleshooting

Argument and upstream failures come back as tool results carrying the same error envelope as HTTP: invalid_request, unsupported_geo, unsupported_parameter, upstream_error, upstream_timeout. The two exceptions are 401 and 429, which are answered at the HTTP level and so show up in a client as a connection or transport failure rather than as a tool result.

• 401 — the credential is missing or wrong. On claude.ai the header value must start with Bearer ; pasting the bare key fails.
• “Couldn’t reach the MCP server” while signing in — that is OAuth discovery failing. Check that https://api.serpresult.com/.well-known/oauth-protected-resource/mcp loads.
• A connection error right after a burst of searches — that is the 429 rate limit. MCP clients render it as a transport failure, not as a tool error; wait a minute and retry.
• A result starting with unsupported_geo — a bad argument, not an outage. The refusal lists every supported code. See Errors.
• GET on the endpoint returns 405 — deliberate. The transport is POST-only, so a browser opening the URL gets 405 with Allow: POST.

Find it in a directory

SerpResult is being listed in the Official MCP Registry as com.serpresult/serp-search, the Claude Connectors Directory and the ChatGPT app directory; until a listing is live, connect with the URL above.