webref/docs
webref/docs
Back to webref

Quick Start

Every time your AI agent searches the web, it gets back thousands of tokens of noisy JSON — navigation menus, cookie banners, ads, duplicate content across sources. That noise fills up context fast, and by the fifth search your agent is forgetting your original instructions. webref fixes this. It queries the same sources but returns focused, synthesized markdown — the same practical information in roughly 10x fewer tokens. Your agent stays sharp, searches more, and actually gets smarter with each lookup instead of drowning in irrelevant data.

webref runs as a remote MCP server. Connecting it to your agent takes one paste — there is nothing to install on your machine.

1. Sign up

Create an account at webref.ai/signup. You can connect your agent first, then choose a plan from billing when you are ready to run research.

2. Copy your install snippet

Open your dashboard. The install card at the top has a tab for each agent — Claude Code, Cursor, Codex, Gemini CLI, Copilot, OpenCode — plus a Generic tab that works anywhere. Pick your agent and click copy. The snippet already has your API key filled in.

3. Paste it into your agent

Run the dashboard command for Claude Code or Codex in your terminal. For editor-based agents, add the supplied config block. Restart your agent or open a new session after registering the server so it discovers WebRef's tools.

If an older installation tells your agent to run the local webref command, follow the CLI migration guide.

4. Research the web

Ask your agent anything that needs current information:

Use webref to find the current Prisma + Turso setup.

The agent calls the WebRef MCP research tool and gets back a focused, citation-traced answer. If research is still running, it resumes with get_research. One research request replaces a search-then-fetch loop.

What's next

Found an issue? Let us know