Claude web search tool alternatives in 2026

Give Claude SpecStream as a tool

Replace the web_search server tool with a client tool of the same name. Claude asks for a search, your handler calls the SpecStream Search API and returns the answer and the ranked results, and Claude writes its reply from them.

import json, os, requests

ANTHROPIC = "https://api.anthropic.com/v1/messages"
HEADERS = {"x-api-key": os.environ["ANTHROPIC_API_KEY"], "anthropic-version": "2023-06-01"}

web_search = {
    "name": "web_search",
    "description": "Search the live web. Returns the answer with its page, and the ranked results with title, URL and snippet.",
    "input_schema": {"type": "object", "properties": {"q": {"type": "string"}}, "required": ["q"]},
}

def search(q):
    reply = requests.get(
        "https://api.specstream.ai/v1/web-search",
        params={"q": q},
        headers={"Authorization": f"Bearer {os.environ['SPECSTREAM_API_KEY']}"},
    ).json()["data"]
    sources = [{"title": s["title"], "url": s["url"], "snippet": s["snippet"]} for s in reply["sources"]]
    return json.dumps({"answer": reply["answer"], "sources": sources})

messages = [{"role": "user", "content": "Who is the CEO of Mistral AI?"}]
while True:
    response = requests.post(ANTHROPIC, headers=HEADERS, json={
        "model": "claude-opus-5-5", "max_tokens": 1024, "tools": [web_search], "messages": messages,
    }).json()
    messages.append({"role": "assistant", "content": response["content"]})
    if response["stop_reason"] != "tool_use":
        break
    messages.append({"role": "user", "content": [
        {"type": "tool_result", "tool_use_id": block["id"], "content": search(block["input"]["q"])}
        for block in response["content"] if block["type"] == "tool_use"
    ]})

print("".join(block["text"] for block in response["content"] if block["type"] == "text"))

With Anthropic's MCP connector the Claude API calls SpecStream for you. Add the server to mcp_servers with your SpecStream API key as the token, and turn on the search_web tool alone.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: mcp-client-2025-11-20" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-5-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Who is the CEO of Mistral AI?"}],
    "mcp_servers": [{
      "type": "url",
      "url": "https://api.specstream.ai/mcp",
      "name": "specstream",
      "authorization_token": "'"$SPECSTREAM_API_KEY"'"
    }],
    "tools": [{
      "type": "mcp_toolset",
      "mcp_server_name": "specstream",
      "default_config": {"enabled": false},
      "configs": {"search_web": {"enabled": true}}
    }]
  }'

Your model, prompt and conversation stay as they are

Claude decides when to search, reads the results and writes the answer with the pages it used, as it does with the built-in tool. Your handler decides what reaches the context: the answer and ten titles, URLs and snippets, or the page text with passages=true.

$5 per 1,000 searches, with no tokens billed by SpecStream

100,000 searches cost $500 on SpecStream. On the Claude web search tool they cost $1,000, plus the token costs for the content each search brings back, which count again in every later turn. Each SpecStream reply carries price_usd, and the account's request log and usage totals show every call and its price. A new account gets $10 of credit, no card, which covers 2,000 searches.

The most accurate web search API in the world

The SpecStream Search API scores a 97.9 F-score on all 1,000 official SimpleQA Verified questions, with 978 correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks, 100 in English and 100 in Chinese, in our run of November 8, 2026. The scores and the method are at specstream.ai/benchmarks.

Results your code can read and keep

The Claude web search tool returns each page's content as encrypted_content, which the API decrypts in later turns and your code cannot read. A SpecStream search returns plain JSON: the answer, the page it was read on and the time it was read, the rows and the ranked sources. The same search serves Claude, GPT, Gemini, a local model or a script with no model.

const reply = await fetch(
  "https://api.specstream.ai/v1/web-search?" + new URLSearchParams({ q: "Who is the CEO of Mistral AI?" }),
  { headers: { Authorization: `Bearer ${process.env.SPECSTREAM_API_KEY}` } },
).then((r) => r.json());

const { answer, sources, price_usd } = reply.data;
console.log(answer?.value, answer?.page, price_usd);

Every charged search is kept in the account's history at GET /v1/web-search/answers, and reading it costs nothing. Turn on zero_retention on a key and its searches keep no request text and no answer after the reply.

Page text, and whole datasets

The Extract API returns the main text of up to 20 pages or documents a call at $0.80 per 1,000 pages, and a page with no text is not charged. Deep Search takes a request in plain words and returns every match on the live web as structured rows at 5¢, 15¢ or $1 a row, each value with its source URL, under a maximum you approve before it runs.

SpecStream

The live web data engine for teams and agents, and the best alternative to the Claude web search tool. The Search API costs $5 per 1,000 searches, bills no tokens, works with Claude and every other model, and scores 97.9 F-score on all 1,000 official SimpleQA Verified questions. The Extract API costs $0.80 per 1,000 pages.

OpenAI web search tool

A tool inside a response from an OpenAI model, at $10 per 1,000 calls plus search content tokens at the model's rates, or $25 per 1,000 calls for web search preview on other models. It serves OpenAI models. SpecStream costs $5 per 1,000 searches and serves Claude and OpenAI models from the same key.

Gemini grounding with Google Search

Grounding on the Gemini API. On Gemini 3 and newer, 5,000 search requests a month are free, then $14 per 1,000, billed for each query the model decides to run, with a default limit of 1,500 requests a day on the paid tier. On Gemini 2.5 it costs $35 per 1,000 grounded prompts. On SpecStream your code decides how many searches run, at $5 per 1,000.

xAI web search

The web_search tool on the xAI API costs $5 per 1,000 calls, and xAI bills the token usage of the request beside it. It filters to at most 5 domains. SpecStream costs $5 per 1,000 searches with no tokens, and returns the answer, its page and the ranked results to Claude or any other model.