Google search API for AI agents
Keep the call, change the address.
An agent has three Google search APIs in 2026. Custom Search JSON is closed to new customers and ends on January 1, 2027. Vertex AI Search covers up to 50 domains, and Gemini grounding returns a model answer in place of results.
SpecStream answers Google's call. Point the agent at https://api.specstream.ai/customsearch/v1 with a SpecStream key in key=, or connect it to the MCP server at https://api.specstream.ai/mcp, and it searches the whole live web.
97.9 F-score on all 1,000 SimpleQA Verified questions and 64.0 item F1 on all 200 WideSearch tasks, the top published scores, on the benchmarks page. $5 per 1,000 searches.
New accounts get $10 of credit, no card. It covers 2,000 searches.
The Google search APIs an agent can call, next to SpecStream
| SpecStream | Custom Search JSON API | Vertex AI Search (Agent Search) | Gemini grounding with Google Search | |
|---|---|---|---|---|
| Open to a new agent | Yes. A key takes a minute and starts with $10 of credit, no card | No. Closed to new customers. Existing customers have until January 1, 2027 | Yes, in a Google Cloud project, with 10,000 free queries per account a month on the trial | Yes, in the Gemini API, with 5,000 free search requests a month on Gemini 3 and newer |
| How the agent calls it | Google's GET /customsearch/v1 at https://api.specstream.ai, GET /v1/web-search, or the search_web tool on the MCP server | GET /customsearch/v1 with a key and the cx of a Programmable Search Engine | POST to a serving config of an engine in your Google Cloud project | A tool inside a Gemini model request. The model decides which queries to run |
| What it searches | The whole live web, read at the moment of the call | The Programmable Search Engine you configured | Up to 50 domains. For full web search Google asks you to contact it | Google Search, through the queries the model runs |
| What comes back | On the Google route, Google's items[]. On /v1/web-search, the answer with the page it was read from and its read time, rows, every source read and, with passages=true, the text of the top pages | An items array with the URL, title and text snippets of each result | Semantic retrieval on Standard. Enterprise adds generative answers | A model response grounded in the search results, with search suggestions as HTML to render |
| Price per 1,000 | $5 for searches of up to 10 results | $5 per 1,000 queries after 100 free a day, for existing customers | $1.50 per 1,000 queries on Standard, $4.00 on Enterprise | $14 per 1,000 search requests after the free 5,000 |
| What one unit is | One call of up to 10 results, $0.005. Each started block of 10 more results adds $0.006 | One query, up to 10,000 a day | Any API call to Enterprise Search, direct or through an integration or the console | Each search query the model decides to run, so one request can be several |
Why agents search through SpecStream
The call your agent already sends
# Before
curl "https://www.googleapis.com/customsearch/v1?key=$GOOGLE_KEY&cx=$CX&q=who+founded+anthropic"
# After
curl "https://api.specstream.ai/customsearch/v1?key=$SPECSTREAM_KEY&cx=$CX&q=who+founded+anthropic"The reply is Google's JSON: kind, items[] with title, link and snippet, searchInformation and queries.nextPage. The tool your agent calls and the parser behind it stay as they are.
The top published scores
On SimpleQA Verified the SpecStream Search API scores 97.9 F-score on all 1,000 official questions, 978 of them correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks. The Extract API scores 0.908 F1 on WCXB. Every score and its method is on the benchmarks page.
Fast enough for a loop of tool calls
The Google route runs the fast search. On production on November 9, 2026 it had a median of 0.82 s at 1 search a second and 1.07 s at 20 a second, with the 95th percentile between 1.09 and 1.73 s, and answered 718 of 720 calls. One account takes up to 8,800 searches a minute; a call over that gets a 429 with Retry-After and is not charged.
Results, or the answer with its page
Gemini grounding hands the agent a model's answer. SpecStream gives it the results, and on GET /v1/web-search the answer with the page it was read from and its read time, so the agent cites the page and your model writes the reply. One search returns up to 100 results with num.
$5 per 1,000, on a key with its own limits
A search of up to 10 results costs $0.005, with no token charge and no daily limit. Each key carries its scopes and a spending limit, and zero_retention keeps no request text and no answer after the reply. Agents connect to https://api.specstream.ai/mcp with OAuth or an API key.
Pages and full datasets on the same key
The Extract API reads pages, PDF, Word and Excel files at $0.80 per 1,000 pages. Deep Search returns every match on the live web as structured rows at 5¢, 15¢ or $1 a row, with the most a run can cost approved before it starts.
The Google search APIs, one by one
Custom Search JSON API
Results from a Programmable Search Engine you configured, at $5 per 1,000 queries after 100 free a day, up to 10,000 a day. It is closed to new customers, and Google discontinues it on January 1, 2027. SpecStream answers the same call at its own address, on the whole live web, with no daily limit.
Vertex AI Search (Agent Search)
Google's named alternative for searching up to 50 domains, at $1.50 per 1,000 queries on Standard and $4.00 on Enterprise. For full web search, Google's page asks you to contact it. SpecStream searches the whole live web from the first call at $5 per 1,000.
Gemini grounding with Google Search
A tool inside a Gemini request: the model runs Google searches and returns a grounded answer with search suggestions. Gemini 3 and newer bill each query the model runs, at $14 per 1,000 after 5,000 free a month. SpecStream works with any model, through the API or the search_web tool on its MCP server, at $5 per 1,000.
How SpecStream scores
The same numbers as the benchmarks page, from the official tasks and scorers.
How to give your agent Google's call on SpecStream
1. Create a key under Agents & API in the workspace. A new account gets $10 of credit, no card, which covers 2,000 searches.
2. Set the base address of the agent's search tool to https://api.specstream.ai and put the SpecStream key in key=. q, num and start work as on Google, and cx is accepted and not used.
import requests
def google_search(query: str) -> list[dict]:
reply = requests.get(
"https://api.specstream.ai/customsearch/v1", # was https://www.googleapis.com
params={"key": SPECSTREAM_KEY, "q": query, "num": 10},
timeout=30,
).json()
return [{"title": i["title"], "url": i["link"], "snippet": i["snippet"]} for i in reply["items"]]const reply = await fetch(
`https://api.specstream.ai/customsearch/v1?key=${SPECSTREAM_KEY}&q=${encodeURIComponent(query)}&num=10`
).then((r) => r.json());
const results = reply.items.map((i) => ({ title: i.title, url: i.link, snippet: i.snippet }));3. For an MCP client, add the server https://api.specstream.ai/mcp with OAuth or the API key. The search_web tool takes {"q": "<the question>"} and is billed as one search.
Frequently asked questions
Something we didn't answer? Talk to a human
Does Google have a search API for AI agents?
What is the best Google search API alternative for AI agents?
GET /customsearch/v1 call in Google's JSON, at $5 per 1,000 searches, with a median of 0.82 s on production.Is the Google Custom Search JSON API shutting down?
How much does Gemini grounding with Google Search cost?
How much does Vertex AI Search cost?
Can my agent use SpecStream through MCP?
https://api.specstream.ai/mcp with OAuth or an API key, and the search_web tool runs a search at $5 per 1,000. The agent spends inside the limits you set on the key.Sources
SpecStream wrote this page. Each Google figure comes from Google's own page.
- Google Custom Search JSON API overviewdevelopers.google.com/custom-search/v1/overview, read Nov 10, 2026
- Google Custom Search JSON API, using RESTdevelopers.google.com/custom-search/v1/using_rest, read Nov 10, 2026
- Agent Search pricing, Google Cloudcloud.google.com/generative-ai-app-builder/pricing, read Nov 10, 2026
- Agent Search REST reference, servingConfigs.searchcloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/projects.locations.collections.engines.servingConfigs/search, read Nov 10, 2026
- Grounding with Google Search, Gemini APIai.google.dev/gemini-api/docs/google-search, read Nov 10, 2026
- SpecStream benchmarksspecstream.ai/benchmarks, read Nov 10, 2026
- SpecStream for agentsspecstream.ai/agents.md, read Nov 10, 2026




