next-geo

Serve markdown to LLMs instead of HTML

An autonomous codebase built by the Continue Software Factory

---

Why?

When an LLM visits your site, it gets this:

<div class="pricing-hero_container__x7z2k">
  <div class="pricing-hero_gradient__m4k9p">
    <h1 class="pricing-hero_title__j2x8n">
      <span class="text-highlight_wrapper__k9m2x">Pricing</span>
    </h1>
    <div class="pricing-card_grid__p3k7m">
      <div class="pricing-card_card__n8x2k" data-tier="free">
        <!-- 200 more lines of div soup -->

With next-geo, it gets this:

# Pricing

## Free
Get started at no cost.
- Up to 1,000 requests/month
- Community support

## Pro — $20/month
For growing teams.
- Unlimited requests
- Priority support
- Custom domains

Drop page.md files next to your page.tsx files and they're served automatically when an LLM requests your pages. When no page.md exists, pages are auto-converted from HTML to markdown on the fly.

This is Generative Engine Optimization — making your site readable by the AI agents and LLMs that increasingly drive traffic and surface your content.

Table of Contents

• Quick start
• How it works
• Auto-conversion
• Configuration
• How detection works
• llms.txt
• Discovery Headers
• Caching & Security
• Utilities
• Redirects
• Route groups
• Contributing
• License

Quick start

The easiest way to set up is with a coding agent. Install the skill and let it handle the wiring:

npx skills add continuedev/skills --skill setup-geo

Then ask your agent: "Set up next-geo in this project."

Manual setup

npm install next-geo

Three files to create/update — see Configuration below for details on each:

1. Wrap your config — add withGeo() to next.config.js
2. Add middleware — create middleware.ts with createGeoMiddleware()
3. Add route handler — create app/api/geo/route.ts with createGeoHandler()

Then write page.md files next to any page.tsx:

app/
  pricing/
    page.tsx    ← what browsers see
    page.md     ← what LLMs see

Test it:

curl -H "Accept: text/markdown" http://localhost:3000/pricing
# or
curl http://localhost:3000/pricing.md

How it works

LLMs and AI agents signal they want markdown in three ways:

1. Accept header — Accept: text/markdown (used by Claude Code, Cloudflare Markdown for Agents, etc.)
2. URL suffix — Append .md to any page URL (e.g., example.com/pricing.md)
3. User-Agent — Known bot identifiers (ChatGPT-User, ClaudeBot, GPTBot, PerplexityBot, etc.)

When detected, requests are rewritten to an internal handler that serves your page.md or auto-converts the HTML.

Auto-conversion

Pages without a page.md file are automatically converted from HTML to clean markdown using Turndown. The converter:

• Extracts <main> or <article> content (skips navigation, headers, footers)
• Strips scripts, styles, and decorative elements
• Preserves headings, links, lists, code blocks, and tables
• Adds the page title as an # h1 if not already present

To disable auto-conversion:

createGeoHandler({ enableAutoConversion: false });

Configuration

Middleware options

createGeoMiddleware({
  // Paths to skip (default: /api/*, /trpc/*, /_next/*, /monitoring*)
  excludePaths: ["/api/*", "/admin/*"],

  // Additional User-Agent patterns to treat as LLM bots
  additionalBotUserAgents: ["MyCustomBot"],

  // Disable the .md URL suffix convention (default: true)
  enableMdSuffix: false,
});

Handler options

createGeoHandler({
  // Disable auto-conversion fallback (default: true)
  enableAutoConversion: false,

  // Base URL for internal fetches during auto-conversion
  internalBaseUrl: "http://localhost:3000",

  // Add context for LLMs at the top/bottom of every response
  preamble: (path) => `> Markdown for ${path}. Visit the full page for interactive content.`,
  postamble: "---\nPowered by next-geo",
});

How detection works

Signal	Example	Priority
.md suffix	GET /pricing.md	Highest (explicit)
Accept header	Accept: text/markdown, text/html	Medium (standard)
User-Agent	ClaudeBot/1.0	Lowest (fallback)

The middleware checks signals in priority order and rewrites the first match. Normal browser requests are never affected.

llms.txt

Generate /llms.txt and /llms-full.txt following the llmstxt.org spec.

// app/llms.txt/route.ts
import { createLlmsTxtHandler } from "next-geo/llms-txt";

export const { GET } = createLlmsTxtHandler({
  siteTitle: "My Site",
  siteDescription: "What my site does",
  baseUrl: "https://example.com",
  entries: [
    { path: "/docs", title: "Documentation", description: "API reference and guides" },
  ],
});

// app/llms-full.txt/route.ts
import { createLlmsFullTxtHandler } from "next-geo/llms-txt";

export const { GET } = createLlmsFullTxtHandler({
  siteTitle: "My Site",
  baseUrl: "https://example.com",
});

The handlers auto-discover page.md files in your app directory, extract titles from # headings and descriptions from first paragraphs, and handle Next.js route groups (stripping (group-name) prefixes from URLs).

Discovery Headers

Advertise markdown availability via Link headers (RFC 8288):

import { addDiscoveryHeaders } from "next-geo/discovery";

// In your middleware or response handler:
addDiscoveryHeaders(response, pathname);
// Sets: Link: </pricing.md>; rel="alternate"; type="text/markdown"

Caching & Security

• Static page.md responses: Cache-Control: public, max-age=300, s-maxage=3600
• Auto-converted responses: Cache-Control: private, max-age=300 (cookies are forwarded, so response may be user-specific)
• Vary: Accept is added globally via the config wrapper so CDNs maintain separate caches
• Vary: User-Agent is added only to markdown responses
• Auto-conversion guards against SSRF (same-host validation, same-host-only redirect following, size limits)

Utilities

You can also use the detection logic directly:

import { isLlmRequest, detectLlmSignal } from "next-geo";

// Simple boolean check
if (isLlmRequest(pathname, acceptHeader, userAgent)) {
  // ...
}

// Detailed signal info
const signal = detectLlmSignal(pathname, acceptHeader, userAgent);
// { type: 'accept-header', quality: 1.0 }
// { type: 'user-agent', botName: 'ClaudeBot' }
// { type: 'md-suffix', originalPath: '/pricing' }
// null

Redirects

When multiple routes render the same content (e.g., / and /home both show the homepage), you can use a frontmatter redirect in page.md to serve another path's markdown instead of duplicating it:

---
redirect: /
---

When the handler finds a redirect field, it resolves the target path's markdown (including its page.md or auto-conversion) and serves that. Redirect chains are followed up to 3 hops. Any frontmatter is stripped from served content — only the markdown body is returned.

This is useful for aliased routes, vanity URLs, or any case where two paths should serve identical LLM content.

Route groups

Next.js route groups (e.g., (main-layout), (marketing)) are handled automatically. The handler scans all (...) directories at the app root to find page.md files regardless of which route group they belong to.

Contributing

We welcome contributions! See CONTRIBUTING.md for development setup, project structure, and the PR process.

License

Apache-2.0 — see LICENSE for details. Copyright (c) 2025 Continue Dev, Inc.

---

Built by Continue