Cloudflare’s Markdown for Agents, announced on February 12, 2026, lets AI agents and crawlers request a web page as Markdown ( Cloudflare introduced Markdown for Agents, a feature that (when enabled for a Cloudflare zone) allows clients such as AI agents to send Cloudflare frames this as a token-efficiency and preprocessing problem: if you feed raw HTML into an LLM pipeline, you pay for navigation, wrappers, scripts, and other markup that has little semantic value. As a concrete example, Cloudflare says the blog post you’re reading takes 16,180 tokens as HTML but 3,150 tokens as Markdown—an 80% reduction. Bottom line: by pushing HTML → Markdown conversion to the edge, agents can spend less effort on extraction/cleanup and reduce both token cost and implementation complexity. text/markdown) instead of HTML. The key idea is simplicity: the publisher turns it on in Cloudflare, and the client just sends an Accept header indicating it prefers Markdown.
What did Cloudflare announce?
Accept: text/markdown. Cloudflare then fetches the original HTML from the origin, automatically converts it to Markdown when possible, and returns Markdown to the requester.
How it works (key mechanics)
Use Accept to request the format
This is standard HTTP content negotiation. If the client includes text/markdown in the Accept header, Cloudflare retrieves the HTML from the origin and returns a Markdown representation when conversion is supported. The response includes Vary: accept, signaling that caches should treat HTML vs. Markdown as different variants.
x-markdown-tokens
Converted responses include an x-markdown-tokens header containing an estimated token count for the Markdown body. In agent implementations, this is useful for context-window budgeting and deciding chunk sizes before you do any downstream processing.
Content-Signal header
Cloudflare’s documentation notes that, by default, converted responses include Content-Signal: ai-train=yes, search=yes, ai-input=yes. This is part of the broader Content Signals approach—an HTTP-level way to communicate preferences about how content may be used after access (with mention of future custom policy options).
How to use it
Try it with curl
If Markdown for Agents is enabled on the target zone, request Markdown like this:
curl https://developers.cloudflare.com/fundamentals/reference/markdown-for-agents/ \
-H "Accept: text/markdown"Implement it with fetch
The client-side logic is equally simple in Workers or any HTTP client—just send an Accept header (conceptual example below):
const r = await fetch("https://developers.cloudflare.com/fundamentals/reference/markdown-for-agents/", {
headers: {
Accept: "text/markdown, text/html",
},
});
const tokenCount = r.headers.get("x-markdown-tokens");
const markdown = await r.text();
Note: This only works if the publisher has enabled Markdown for Agents on their Cloudflare zone. It’s not a universal “convert any site on demand” switch—it’s negotiation with sites that have opted in.
How to enable it
Dashboard
Cloudflare’s docs describe enabling this per zone in the Cloudflare dashboard (plan requirements apply). In practice, you select the zone and toggle the Markdown for Agents setting within the relevant AI/Crawler controls area (Cloudflare also highlights Quick Actions in its documentation).
Enable via API
Via the API, you patch the zone setting at /client/v4/zones/{zone_tag}/settings/content_converter with {"value": "on"}:
curl -X PATCH "https://api.cloudflare.com/client/v4/zones/{zone_tag}/settings/content_converter" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer {api_token}" \
--data-raw '{"value": "on"}'
Pricing note: Cloudflare states the feature is available at no extra cost for Pro, Business, and Enterprise plans, and also for SSL for SaaS customers.
Limitations and operational gotchas
HTML only (for now)
Per Cloudflare’s current limitations, conversion is limited to HTML → Markdown. Don’t assume the same mechanism will convert PDFs or Office documents.
No compressed origin responses
Cloudflare explicitly notes that it doesn’t support compressed responses from the origin for this conversion flow. Depending on how your origin is configured, this can be a surprising reason the feature doesn’t behave as expected.
Zone-level setting
This is configured at the zone level. If you need different behavior per subdomain, Cloudflare’s guidance is to split them into separate zones.
Alternatives (and when to use them)
Cloudflare also points to other options for broader “document-to-Markdown” needs. In practice, you’ll pick based on whether you control the zone, what formats you need, and whether you must render JavaScript-heavy pages.
| Option | Best for | Strengths | Trade-offs |
|---|---|---|---|
| Markdown for Agents | HTML from zones where the publisher enabled it | Just send Accept; conversion happens at the edge |
Requires publisher opt-in; has clear limitations |
Workers AI toMarkdown |
Multiple file formats (conversion utility) | Works for documents outside Cloudflare zones; supports many formats | You need to integrate an API; costs can vary by use case |
Browser Rendering /markdown |
Dynamic pages that require real rendering | Converts after browser rendering, which helps with JS-heavy sites | Higher latency and compute cost due to rendering |
Practical use cases
Reduce agent-side preprocessing
Many scraping/agent pipelines end up doing: HTML fetch → main-content extraction → HTML-to-Markdown conversion → chunking. Markdown for Agents offloads the conversion step to the delivery layer, which can simplify your implementation and reduce token waste.
Monitoring and visibility
Cloudflare also notes that Cloudflare Radar can help visualize content types returned to AI bots/crawlers. Operationally, this makes it easier to answer: “Are agents actually consuming our Markdown variant, and how often?”
Summary
Cloudflare’s Markdown for Agents uses Accept: text/markdown to trigger on-the-fly conversion from HTML to Markdown at the edge. Done right, it reduces token waste and removes a common preprocessing step from agent pipelines. The trade-off is that it only works for zones where the publisher enabled it—and today it’s limited to HTML, doesn’t support compressed origin responses, and is configured per zone. If your site is already on Cloudflare, enabling it on a target zone and validating real agent requests is a practical first step.
