Free Remote MCP Server for Web Search: Hosted, No Install, No API Key
Most MCP servers are local programs. Your client starts them with npx or uvx, talks to them over stdio, and you need Node or Python installed on the same machine. That works well in an editor on your laptop. It doesn't work in a browser chat, on a locked-down work machine, or anywhere you can't run a process.
A remote MCP server fixes that. It lives at a URL and speaks the MCP Streamable HTTP transport, so the client just sends HTTP requests. Nothing to install, nothing to keep updated.
AgentSearch now runs a free hosted remote MCP server for web search at:
https://relay.agentsearchhq.com/mcp
No install, no Node, no API key, no wallet and no sign-in. It gives your agent three tools: web search (25 free searches per IP in total, then pay per call), a free trial of URL-to-markdown extract (5 a day per IP) and a free trial of JavaScript render (3 a day per IP). It's listed in the official MCP Registry as com.agentsearchhq/search (remote) and as a Glama connector.
What the hosted server gives you
| Tool | What it does | Free limit (per IP) |
|---|---|---|
agentsearch_web_search | Searches the web and returns hits (title, URL, snippet) as JSON. max_results 1-5, default 5. | 25 searches in total, then pay per call. This allowance does not reset. |
agentsearch_extract | Fetches one URL and returns its title, clean markdown and/or text, links and metadata. | Free trial: 5 calls a day |
agentsearch_render | Loads one URL in headless Chromium so JavaScript runs first, then returns the content (text output only, no screenshots). | Free trial: 3 calls a day |
All three are read-only. They don't click, submit or change anything. The server tells the model when to use which: search when you have a question, extract when you already have a URL, render when extract comes back empty because the page is built with JavaScript. When a limit is reached, the tool returns an error with the paid options instead of failing silently.
The hosted endpoint uses the same free relay and the same per-IP limits as the local package. It's stateless (no session to keep alive) and needs no OAuth. If a client asks about authentication, choose none.
Connect from Cursor and VS Code (one click)
The MCP page has one-click buttons: Add to Cursor (hosted) and Add to VS Code (hosted). If you prefer config files:
Cursor (.cursor/mcp.json):
{
"mcpServers": {
"agentsearch": { "url": "https://relay.agentsearchhq.com/mcp" }
}
}
VS Code (.vscode/mcp.json):
{
"servers": {
"agentsearch": { "type": "http", "url": "https://relay.agentsearchhq.com/mcp" }
}
}
Connect from Claude, Claude Code and ChatGPT
Claude.ai and Claude Desktop (custom connector):
- Open Connectors (Customize > Connectors).
- Select Add, then Add custom connector.
- Name:
AgentSearch. Remote MCP server URL:https://relay.agentsearchhq.com/mcp. Click Continue. - Authentication: no sign-in. Click Add.
This works on Free (1 custom connector), Pro, Max, Ultra, Team and Enterprise (an Owner adds it).
Claude Code:
claude mcp add --transport http agentsearch https://relay.agentsearchhq.com/mcp
ChatGPT (custom MCP server, web):
- Go to chatgpt.com/plugins.
- Select +, then Add custom MCP server.
- Name:
AgentSearch. Server URL:https://relay.agentsearchhq.com/mcp. - Authentication: No authentication.
- Accept the risk warning, select Create as a plugin, install it, then type @ in a chat to use it.
Your plan and workspace permissions decide whether custom MCP servers are available in ChatGPT. Menu labels can change between app versions; the MCP page links the official setup docs for each client.
Any other Streamable HTTP client
Any MCP client or agent framework that supports the Streamable HTTP transport can use the same URL. Many accept a config shaped like this:
{
"mcpServers": {
"agentsearch": { "url": "https://relay.agentsearchhq.com/mcp" }
}
}
Check your framework's docs for the exact key name (some call it url, some serverUrl, some want "type": "http"). Calling AgentSearch from your own code instead of an MCP client? See Use AgentSearch from your own agent.
Test it with curl (no client needed)
Because the server is plain HTTP, you can talk to it with curl. Send the MCP initialize request with an Accept header that allows both JSON and event streams:
curl -s -X POST https://relay.agentsearchhq.com/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"1.0"}}}'
The reply names the server (agentsearch-remote) and includes short instructions for the model. Then list the tools. The server is stateless, so there's no session ID to pass along:
curl -s -X POST https://relay.agentsearchhq.com/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'
You'll get back agentsearch_web_search, agentsearch_extract and agentsearch_render with their full input schemas. Listing tools doesn't use any of your free searches. A real search is a tools/call:
curl -s -X POST https://relay.agentsearchhq.com/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"agentsearch_web_search","arguments":{"query":"model context protocol streamable http","max_results":3}}}'
That call counts as one of your 25 free searches.
When to switch: the local package with a wallet, or direct x402
The hosted server is the fastest way to try AgentSearch and is enough for light use. Once your agent needs more than the free allowance, you have two pay-per-call options, both with no account, no API key and no minimum:
- The local MCP package with a wallet. Run
npx -y @agentsearchhq/agentsearch-mcpover stdio and add a dedicated wallet key with a spend cap. SetAGENTSEARCH_MODE=paidto pay for search too. The install section has the config and wallet setup walks through creating a small, separate wallet. - The direct x402 API at
https://api.agentsearchhq.com, called from your own code. Each call is paid in USDC on Base: search $0.003, extract $0.002, render $0.005, search + read $0.005, news $0.003 and screenshot $0.004. Calls are settled only after a successful answer. See the x402 docs and the pricing page.
Rule of thumb: stay on the hosted URL while you prototype or for a chat client that can't run local servers. Move to the wallet-backed package when an editor agent searches all day, and to direct x402 when your own backend makes the calls.
FAQ
Is the hosted MCP server really free?
Yes, within its limits: 25 free searches per IP in total, then pay per call, plus free trials of 5 extracts and 3 renders a day per IP. No key, wallet or sign-in is needed to use those.
Does the 25-search allowance reset?
No. It's a lifetime allowance per IP, shared with the free relay. The extract and render trials are daily.
Do I need to install anything?
No. You only need an MCP client that supports remote servers over Streamable HTTP. No Node.js, Python or Docker.
Does it use OAuth or an API key?
Neither. There is no authentication. If your client asks, choose no authentication.
What is the difference from the npx package?
The hosted server has the free tools only. The local @agentsearchhq/agentsearch-mcp package runs on your machine and can pay per call from a wallet you control, with a spend cap, so it keeps working past the free limits.
Can a web page tell my agent what to do?
Treat everything extract and render return as untrusted third-party data, as the tool descriptions say. Don't let page text trigger actions your agent wouldn't take otherwise.
Get started
Paste https://relay.agentsearchhq.com/mcp into any client that takes a remote MCP URL, or use the one-click buttons on the MCP page. For client-specific walkthroughs, see How to Add Free Web Search to Cursor with MCP, How to Add Free Web Search to Claude Code with MCP, How to Add Free Web Search to OpenAI Codex CLI with MCP and Free MCP Web Scraping for AI Agents.
More from the blog
- How to Add Free Web Search to OpenAI Codex CLI with MCP (No API Key)
- How to Add Free Web Search to Cursor with MCP (No API Key)
- How to Add Free Web Search to Claude Code with MCP (No API Key)
- How to Add Free Web Search to VS Code Copilot Agent Mode with MCP (No API Key)
- Free MCP Web Scraping for AI Agents: Search, Extract and Render With No API Key
Service pages: Web Search API · Web Extract API · Web Render API