Which web search API should I connect to my n8n or OpenClaw agent?
For most n8n and OpenClaw agents, Brave or Tavily is the simplest web search API to connect: both have a native integration on both platforms and return results a model can read in 1 to 2 seconds. Connect a Google SERP A
For most n8n and OpenClaw agents, Brave or Tavily is the simplest web search API to connect: both have a native integration on both platforms and return results a model can read in 1 to 2 seconds. Connect a Google SERP API instead when the agent must see Google's ranking or its AI Overview, and accept about 8 seconds per search.
-
Native support: OpenClaw's
web_searchtool supports 16 providers and detects Brave first. n8n has verified nodes from Brave, Tavily, Exa, SerpApi and cloro, and a built-in SearXNG tool. - Free to start: OpenClaw has keyless providers (DuckDuckGo, Parallel Search Free, SearXNG). Each paid API gives a small monthly allowance, listed in the table below.
-
Search is half the job: a search API returns links and snippets. To read a page, add n8n's HTTP Request tool or OpenClaw's
web_fetch, or pick an API that returns page text.
The steps, in short:
- Pick the API from the table below by what the agent must answer.
- Get a key from that vendor's dashboard.
- In n8n, attach the vendor's node to the AI Agent's Tool input. In OpenClaw, run
openclaw configure --section webor put the key in~/.openclaw/.env. - Ask the agent one question that needs today's web, and check that it calls the tool.
Which search APIs work natively in n8n and OpenClaw?
| API | n8n | OpenClaw | Median search time | Live price per 1,000 | Free allowance |
|---|---|---|---|---|---|
| Brave | Verified node, incl. LLM Context | Built-in provider, detected first | 1.21 s | $5 | $5 of credit a month |
| Tavily | Verified node: Search, Extract, Crawl, Map, Research | Built-in provider | 2.00 s | $8 pay as you go | 1,000 credits a month |
| Exa | Verified node | Built-in provider | 0.74 s | $4 Lite, $7 Auto | $10 of credit a month |
| Parallel | Not checked | Built-in provider, plus a keyless free provider | 2.99 s | $1 fast, $5 advanced | Keyless provider in OpenClaw |
| SerpApi | Verified "SerpApi Official" node | No provider; connect over MCP or HTTP | 8.20 s | $25 on Starter | 250 searches a month |
| cloro | Verified node | Plugin or MCP | 7.96 s | $1.20 on Hobby ($100 for 250,000 credits) | 500 credits a month |
| SearXNG | Built-in tool node | Built-in keyless provider | Not measured | Free, self-hosted | No limit set by a vendor |
| DuckDuckGo | Not checked | Built-in keyless provider | Not measured | Free | Not published |
Median times come from an open benchmark of six web search APIs in November 2026. Integration facts come from n8n's integration pages and the OpenClaw web tools docs, read on November 16, 2026.
Two notes on the n8n column. The older built-in "SerpApi (Google Search)" tool is deprecated as of n8n 2.35.0, and the docs point to the verified SerpApi Official node. And n8n's Perplexity node sends chat messages to a Perplexity model; it is not a search API node.
How do you pick one?
| Your agent must | Connect |
|---|---|
| Answer a person in a chat, fast | Brave or Exa |
| Read whole pages, not only snippets | Tavily (Extract) or Exa (contents), or any API plus a fetch step |
| Run at no cost on your own server | SearXNG, or DuckDuckGo in OpenClaw |
| See Google's ranking, AI Overview or People Also Ask | A Google SERP API: SerpApi or cloro |
| Answer hard, multi-hop questions in the background | A Google SERP API; in the benchmark they answered 77% to 80% of multi-hop questions, against 48% to 65% for the others |
The n8n forum thread that asks how to let an agent search the web gets the same answer from the other side: a SERP tool returns results for keywords, and the agent then needs the HTTP Request tool to read the pages. Plan for that second tool, or for an API that returns page text, before you pick on price.
How do you connect a search API to an n8n AI Agent?
- Install the node. In the nodes panel, search for the vendor's name and pick the version with the Verified badge. An instance owner must set up verified nodes once before others can install them.
- Create the credential. Paste the API key from the vendor's dashboard.
- Attach it as a tool. On the AI Agent node, click the plus sign under Tool and add the vendor's node. Write a one-line description such as "Search the web for current facts", so the model knows when to call it.
- Add a page reader. Attach the HTTP Request tool if the agent must open the result URLs.
- Test with a question about this week. The model will answer old questions from memory and skip the tool.
If a verified node does not show under Tool on a self-hosted instance, n8n forum threads report that community nodes need N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true in the environment. For an API with no node, the HTTP Request tool can call any JSON endpoint with a header key.
How do you set the web search provider in OpenClaw?
OpenClaw reads the provider from tools.web.search.provider. When that is empty, it checks credentials in a fixed order and uses the first one that is ready: Brave, MiniMax Search, Gemini, Grok, Kimi, Perplexity, Firecrawl, Exa, Tavily, Parallel, DeepSeek, SearXNG. A keyless provider never wins that check, so select it by name.
To get a Brave Search API key for OpenClaw:
- Sign up at
api-dashboard.search.brave.comand choose the Search plan. Brave asks for a card and gives $5 of credit a month. - Create an API key in the dashboard.
- Run
openclaw configure --section web, or addBRAVE_API_KEY=...to~/.openclaw/.envand restart the gateway. - To set the provider by name, add it to the config:
{
tools: {
web: {
search: { provider: "brave", maxResults: 5 },
},
},
}
For Tavily, Exa or Parallel, the steps are the same with TAVILY_API_KEY, EXA_API_KEY or PARALLEL_API_KEY. For no key at all, set provider: "parallel-free", or point SEARXNG_BASE_URL at your own SearXNG instance.
Gemini is the one built-in provider that uses Google Search. It returns one answer with citations through Google Search grounding, and no list of results. For Google's ranked results, the AI Overview or People Also Ask, connect a Google SERP API as an MCP server:
openclaw mcp add google-serp \
--url https://mcp.cloro.dev/YOUR_API_KEY/mcp \
--transport streamable-http
Note the speed: in the benchmark, the two Google SERP APIs took about 8 seconds a search, against 0.74 to 2.99 seconds for the APIs with their own index. That is fine for a background task and slow for a chat.
How this was checked
This article is published by cloro, which sells one of the APIs in the first table and maintains its n8n node. The n8n facts come from n8n's own integration pages and docs, read on November 16, 2026; the tool-usage variable comes from n8n community forum threads and is not in the n8n docs page that was checked. The OpenClaw facts come from the OpenClaw web tools docs, read the same day. Every price comes from the vendor's own pricing page. The speed and accuracy figures come from an open benchmark that publishes every graded answer. A full n8n rank-tracking workflow on a SERP API, with the Code node, is in SERP API With Native n8n and Zapier Integrations.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.