Publish to Cloudflare Pages from Claude: research, write, deploy

Yes, through Claude Code, not through the connector alone. Cloudflare's Pages documentation lists three ways to ship: connect a Git provider for automatic deployments, upload prebuilt assets with Direct Upload, or create a project from the command line. For Direct Upload the documented command is npx wrangler pages deploy followed by the build output directory, and adding a branch flag deploys to a preview environment.

Cloudflare's Claude Code setup page describes the plugin that installs its skills, among them a Wrangler skill, and registers its MCP servers. The Developer Platform connector's listed tools cover Workers, D1, KV, R2 and Hyperdrive, with no Pages deploy among the 24 tools, so Claude reaches for Wrangler or Git for the publish itself. Use the connector to read what your Workers and storage hold, and keep the deploy in Claude Code where you can see each command before it runs.

Git suits a site whose posts live in a repository. A tutorial from TravTeks pushes the site to GitHub, connects Cloudflare Pages to the repo, and gives Claude a fine-grained token limited to that one repository. It says Claude then clones the repo, reads the conventions, edits files, commits and pushes, and that each push goes live in about 60 to 90 seconds, with the previous deploy staying live if a build fails.

Direct Upload suits a site you build locally. Cloudflare says Wrangler can upload up to 20,000 files of 25 MiB or less, and drag and drop is capped at 1,000. It also warns that a Direct Upload project cannot later switch to Git integration. Choose once. Cloudflare's own Pages page also says Workers supports most Pages use cases and recommends starting new projects with Workers, so check that page if you are starting fresh.

A person on Reddit asking how to publish a Claude-made site put it this way: "claude can produce a working single-file page, but sharing the artifact isn't the same as publishing something you can update later." Publishing is half the job. The TravTeks tutorial, which ranks on this query's first page, covers the deploy and does not cover research, measuring or SEO.

The missing half is what Linkeddit adds. Its keyword_research_get lists scored keywords with intent and a suggested landing page, and keyword_serp shows who ranks now. On the test project one research job scored 60 keywords and filtered 42 out, which is how much a first pass removes before anyone writes. Choosing the query first means the post answers something people ask.

Build in small steps as well. A Claude Code user on Reddit who built a site on Cloudflare Workers and D1 wrote that asking for a plan first, then building in phases, "gave me much more control than asking for everything at once." The prompt on this page follows that advice: write, build, read the diff, deploy a preview, and only then promote.

Cloudflare's SEO guidance says to make sure your WAF custom rules, rate limiting rules and bot protection settings are not blocking Verified Bots, and it suggests enabling Crawler Hints so search engines have the freshest version of your content. It also says that using Cloudflare's nameservers and its caching do not affect SEO.

For AI answers, the crawler settings matter separately: Cloudflare's AI Crawl Control lets you allow or block individual crawlers, and a blanket block can keep a new post out of ChatGPT search. Then let Linkeddit's Search and Answer Radar record whether the page ranks and whether the engines name you. Cloudflare's Pages page lists 500 deploys a month on the Free plan, so batch edits instead of deploying after each one.