Skip to content
Pollen

Web data for AI agents, paid forward.

Pollen turns any page into clean Markdown through a Firecrawl-compatible API and an MCP server. We are building a share into every fetch: paid to the site if it has claimed it, otherwise returned to you as model tokens.

Get API KeySee the API
markdown
engine
http (no browser needed)
time
…
words
4,343
links
340

Fair share in development

Once live, the share of this fetch goes to en.wikipedia.org if it has claimed it, otherwise back to you as tokens.

Replays responses our API returned on 2 Nov 2026. Live fetches need an API key.

Built for agents that already read the web: Hermes Agent and OpenClaw through the Firecrawl SDKs, Claude Code, Codex and other MCP clients through our MCP server.

Agents take content. Pollen sends something back.

A bee feeds on a flower and pollinates it in return. We want web fetches to work the same way: part of what you pay for each fetch is set aside for the site you read.

If the site has claimed its share
The publisher verifies the domain once and receives the shares from then on.
If it has not
The share is credited to your account as model tokens, so your agents get cheaper the more they read.

Status: in development. The fetch API runs today; claims, shares and token credits come after the beta.

  1. Your agentcalls /scrape, /crawl or MCP
  2. Pollen fetches and sets the share aside
  3. Claimed siteshare paid to the publisher
  4. Unclaimed siteshare back to you as tokens

The routes and response shape you know from Firecrawl.

Existing SDKs and agents keep working when you change the base URL. Scrape is the route most agents call, so it gets the most care.

POST /v2/scrape

Scrape

A plain HTTP fetch first. A real browser only when the page needs JavaScript, and a remembered choice for sites that always do. You get the main content without menus, cookie banners and footers.

  • Markdown, HTML, links and metadata in one call
  • Clear error codes when a site blocks or rate-limits us
  • TLS certificates are always verified

https://en.wikipedia.org/wiki/Web_scraping

Response
268ms
Words
4,343
Links
340

One fetch through our API, 2 Nov 2026.

Crawl
Walks a site breadth-first as an asynchronous job, within the page, depth and path limits you set. Poll it or cancel it.
POST /v2/crawl
Map
Lists a site's URLs from its sitemaps and page links, so you can choose what to read before you pay for it.
POST /v2/map
Documents
PDFs with a text layer and Word files come back as Markdown with headings and paragraphs.
PDF, DOCX
Screenshots
The viewport or the whole page as JPEG, in the same call as the Markdown.
screenshot@fullPage
MCP server
Gives MCP clients the scrape and map tools, backed by the same API and key.
fair_fetch_scrape, fair_fetch_map

Already on Firecrawl? Change the base URL.

Same requests, same response. The Python SDK, plain HTTP and our MCP server all use the same key.

# pip install firecrawl-py
import os
from firecrawl import Firecrawl

app = Firecrawl(
    api_key=os.environ["POLLEN_API_KEY"],
    api_url="https://fetch.naolhaase.dev",
)

page = app.scrape(
    "https://en.wikipedia.org/wiki/Web_scraping",
    formats=["markdown"],
)
print(page.markdown[:400])

Response, HTTP 200

{
  "success": true,
  "data": {
    "markdown": "# Web scraping\n\n**Web scraping**, …",
    "links": ["…"],
    "metadata": {
      "title": "Web scraping - Wikipedia",
      "statusCode": 200,
      "sourceURL": "https://en.wikipedia.org/wiki/Web_scraping"
    }
  }
}

Measured against Firecrawl on the same pages.

PollenFirecrawl
Main-content accuracy (F1)higher is better
Pollen 0.933
Firecrawl 0.797
Precision: share of output that is real contenthigher is better
Pollen 0.897
Firecrawl 0.676
Markdown size for the same contentsmaller means fewer tokens
Pollen 0.55×
Firecrawl 1.00×
Median fresh fetch, plain HTTP pageslower is better
Pollen 1.35 s
Firecrawl 1.73 s

Measured in November 2026. Accuracy: the same stored HTML of 181 pages from a public article benchmark, scored against hand-labelled main content. Speed: 40 pages fetched one after another, counting only fresh fetches (Firecrawl cache misses).

For publishers. Planned, not live yet.

AI agents already read your site. You should get a share.

You will verify your domain with one DNS record. From then on, the shares from fetches of your pages go to you, and you can see which pages agents read. Your robots.txt and opt-out signals are already recorded on every fetch today.

DNS TXT record (format may change before launch)

_pollen.your-site.com

"claim=token-from-your-dashboard"

Claims open after the beta. Until then, nothing to sign up for here.

Private beta. Keys are issued by hand.

We give out a small number of keys while we calibrate the fair share and watch how the service holds up.

Key requests open soon.