Plainserp is one endpoint that returns Google’s organic results as JSON. You send a query and get back titles, links and snippets. It’s built for AI agents and scripts that need search results and nothing around them.

The base address is https://plainserp.com. Every request is a GET over HTTPS, and every response is JSON.

1. Create a key on the API keys page.
2. Save it as PLAINSERP_KEY in your environment.
3. Make a request.

curl "https://plainserp.com/v1/search?q=postgres+connection+pooling&num=3" \
  -H "Authorization: Bearer $PLAINSERP_KEY"

{
  "query": "postgres connection pooling",
  "results": [
    {
      "position": 1,
      "title": "PgBouncer - lightweight connection pooler for PostgreSQL",
      "url": "https://www.pgbouncer.org/",
      "snippet": "PgBouncer keeps a pool of server connections and hands them to clients…"
    },
    {
      "position": 2,
      "title": "PostgreSQL: Documentation: Connections and Authentication",
      "url": "https://www.postgresql.org/docs/current/runtime-config-connection.html",
      "snippet": "max_connections determines the maximum number of concurrent connections…"
    },
    {
      "position": 3,
      "title": "pgpool Wiki",
      "url": "https://www.pgpool.net/",
      "snippet": "Pgpool-II is a middleware that works between PostgreSQL servers and clients…"
    }
  ],
  "cost_usd": 0.0003
}

Send your key as a Bearer token in the Authorization header. Keys start with ps_live_.

Authorization: Bearer $PLAINSERP_KEY

Keep keys on your server. Anyone who has a key can spend your balance, so revoke a key on the dashboard if it leaks.

GET /v1/search runs one Google search and returns the organic results.

A successful call returns status 200 with this shape.

Google returns results in pages of up to 20, and goes 6 pages deep, which is the first 120 results of a query. Set pages to fetch several pages in one call. They come back as one list, in order.

# Pages 1 to 3: results 1 to 60, charged as three searches
curl "https://plainserp.com/v1/search?q=postgres+connection+pooling&num=20&pages=3" \
  -H "Authorization: Bearer $PLAINSERP_KEY"

Each page is charged as one search, so the call above costs $0.0009. If a page fails it isn’t charged, and you get the pages that worked. To fetch one page from deeper in the results, set page instead, for example page=4.

Errors use standard HTTP status codes and a JSON body with a code you can match on. You are never charged for a request that returns an error.

{
  "error": {
    "code": "insufficient_balance",
    "message": "Your balance is too low. Top up to continue."
  }
}

For 429 and 502, wait a second or two and try again. Doubling the wait after each failed attempt works well.

Each account can make 300 searches a minute by default. The count resets at the start of every clock minute, and a call that fetches several pages counts once per page. Your limit and how much of it you’re using are on the dashboard.

Every response says where you stand. Past the limit you get a 429 with a Retry-After header giving the seconds until the next minute. Requests refused this way are not charged.

X-RateLimit-Limit: 300
X-RateLimit-Remaining: 287
Retry-After: 23

Need more? Message @vvz25 on Telegram and say roughly what you need.

One successful search costs $0.0003, whatever num is. A call that fetches several pages is one search per page. That is $0.30 for a thousand searches. A new account starts with $0.30, which covers your first 1,000 searches.

You add money in advance from $6, and it doesn’t expire. Every response includes cost_usd, and your balance and daily usage are on the dashboard.

Most agent frameworks take a tool definition and a function to run when the model calls it. This definition works as it is with any model that supports tool use.

{
  "name": "web_search",
  "description": "Search Google. Returns titles, URLs and snippets.",
  "input_schema": {
    "type": "object",
    "properties": {
      "q": { "type": "string", "description": "The search query" },
      "num": { "type": "integer", "description": "Results to return, 1 to 20" }
    },
    "required": ["q"]
  }
}

import os, requests

def web_search(q: str, num: int = 10) -> list[dict]:
    r = requests.get(
        "https://plainserp.com/v1/search",
        params={"q": q, "num": num},
        headers={"Authorization": f"Bearer {os.environ['PLAINSERP_KEY']}"},
        timeout=15,
    )
    r.raise_for_status()
    return r.json()["results"]

Asking for 5 to 10 results is usually enough for a model to pick what to read, and it keeps the tool result short.

To give an AI client search without writing any code, connect it to the Plainserp MCP server. It offers one tool, web_search, and uses your API key and balance like any other call.

claude mcp add --transport http plainserp https://plainserp.com/mcp \
  --header "Authorization: Bearer $PLAINSERP_KEY"

{
  "mcpServers": {
    "plainserp": {
      "type": "http",
      "url": "https://plainserp.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

The exact file and field names vary by client, so check your client’s own instructions for adding a remote server with a header.

A skill is a short instruction file that teaches a coding agent how to use something. The Plainserp skill covers the endpoint, parameters, errors, billing and ready-made code, so the agent can add search to your project correctly the first time.

Open the skill file, or install it for Claude Code:

mkdir -p ~/.claude/skills/plainserp-search
curl -o ~/.claude/skills/plainserp-search/SKILL.md \
  https://plainserp.com/skill/plainserp-search/SKILL.md

Then ask your agent to “add web search with Plainserp”. For other agents, put the file wherever that agent reads its instructions from.

• It returns organic results only. AI Overviews, People also ask, knowledge panels, ads and image or video packs are left out.
• It doesn’t fetch or read the pages it links to.
• It reaches the first 120 results of a query and no further.
• Plainserp is an independent service and isn’t affiliated with Google. A change on Google’s side can cause a short outage, and failed requests are never charged.

Questions, a payment that didn’t show up, or a higher rate limit: message @vvz25 on Telegram. Include the email on your account.