Dapper Roadmap
This is the plan for making Dapper the best tool for people who want to build static websites quickly with an AI assistant, and publish them wherever they like.
The idea
Describe the site you want to your AI assistant; Dapper makes it real, keeps it correct, and publishes it wherever you like.
Dapper doesn't contain an AI and never charges for one. It works with the assistant a person already uses (Claude, ChatGPT, Cursor, or anything that speaks MCP or can run a command), and gives that assistant what it needs to do good work safely: clear conventions, guardrails, checks that explain what is wrong, and a fast loop from edit to preview to publish.
Principles
- You own everything. A site is plain files: Markdown, HTML, YAML and CSS. No database, no account, no lock-in. Any host that serves files works.
- Safe by default. Content is never run as code. AI changes are checked
against schemas before they're written, and
dapper checkmust pass before a deploy. - Explain, don't just fail. Every error names the file, says what is wrong, and is available as JSON, so an assistant can fix it without help.
- Bring your own AI. Dapper meets assistants where they are: MCP for
assistants that support it,
AGENTS.mdfor coding agents,--jsonfor scripts. - Small and durable. Dapper has run since 2002. New features earn their place, and a site built today should still build in ten years.
Who it's for
- People who want a site, not a website project: a bakery, a band, a consultant, a club, a portfolio. They use an AI assistant and want the result to look good and stay current.
- Writers and bloggers who want Markdown, a feed and a fast site without running WordPress.
- Developers and documentation teams who want a small, scriptable generator that their agents can drive.
What a session looks like
- Install Dapper in one line (single file, Homebrew or Docker).
dapper init --template business(or a theme from a git URL).- Ask your assistant: "Make this the site for my bakery. Here are our
hours, and here's our menu." It reads
AGENTS.md(or uses Dapper's MCP tools), writes pages that match the schemas, and runsdapper check. dapper serveshows the site; the browser reloads as the assistant edits.dapper deploypublishes only what changed, to S3, GitHub Pages or any host.- Months later: "Post Saturday's event recap with these photos." The assistant knows how, because the project explains itself.
What 1.0 has
Safe content, front matter schemas, data files, pagination,
category and tag pages, a sitemap, an Atom feed, SEO and structured data
tags, Markdown copies and llms.txt, Pagefind search, dapper check --json, the MCP server, clean output, live reload, the blog starter, and
installs as a single file, a Docker image and a Homebrew formula. 1.0 also
makes assistants good at Dapper from the first minute, gives people good
looking starting points, and makes publishing one command:
- AGENTS.md in every project.
dapper initwrites anAGENTS.mdthat explains the project to coding agents: where pages, layouts and data live, the schemas, safe content, and the edit, check, build loop. ACLAUDE.mdpoints Claude Code at it.dapper agentsregenerates it for existing sites. - MCP prompts and resources. Prompts for the common jobs (start a new site, write a post, audit the site) and resources an assistant can read (the Dapper guide, the project file, the schemas).
- More MCP tools.
rename_page(moves a page and adds a redirect from its old URL),read_layoutandwrite_layout(a layout is checked before it is saved). - Redirects.
redirectsin_config.ymlor_data/redirects.ymlmap old URLs to new ones; Dapper writes a small redirect page at each old URL, andcheckmakes sure each target exists. - Themes. Three more starters alongside
blog:portfolio,business(hours, services and contact details from data files) anddocs(a sidebar from a data file). Each is responsive, accessible, has a dark mode, and ships with a schema and anAGENTS.mdthat explains how to fill it in. - Themes from anywhere.
dapper init --templatetakes a built-in name, a local directory or a git URL, so anyone can publish a theme. dapper deploy. Publishes the output directory after a successful build and check:- Amazon S3, through the AWS CLI (
aws s3 sync --delete), with an optional CloudFront invalidation. - GitHub Pages: commits the output to a branch (with
.nojekylland an optionalCNAME) and pushes it, keeping the branch's history. - Any command, such as
rsyncto a server. --dry-runshows what would happen; credentials always come from the standard tools' own configuration, never from_config.yml.
- Amazon S3, through the AWS CLI (
What 1.10 adds
Images (sizes, responsive copies, WebP, lazy loading), components in
Markdown (callouts, details, galleries, videos, buttons, figures, maps, and
components of your own), drafts and scheduled posts, accessibility and SEO
checks in dapper check, and dapper import from WordPress, Jekyll, Hugo
and Medium.
Next
The biggest gap left is making Dapper easy to hand to an AI assistant: to set up, to see what it changed, and to undo.
Working with AI assistants
- One-step setup: list Dapper in the MCP registry, and ship it as a
Claude Code plugin and a Claude Desktop extension, so
dapper mcpis one click away. - Let the assistant see its work: an MCP tool that starts a preview and returns a screenshot of a page.
- Review and undo: every change through MCP becomes a git commit with a summary of what changed in the built site, and undoing the last one is one tool call.
- More MCP tools: add an image (copied, resized, with alt text), delete a
page (with a redirect), and a deploy that runs
checkfirst and asks the owner before publishing.
Quality and content
- Weight checks: oversized images and heavy pages.
- External link checks:
dapper check --externalfollows links to other sites and reports the dead ones. - Pages from data: a page per entry in a data file (a team member, a product, an event), with a layout.
- Themes from HTML templates:
dapper theme importtakes a free HTML template (from HTML5 UP or Start Bootstrap, say) apart into Dapper layouts, keeping its license.
Later
- Forms, comments and analytics as settings: contact forms through
services such as Formspree or Basin, comments through giscus, and
privacy-friendly analytics (Plausible, GoatCounter), each a few lines of
_config.yml. - More deploy targets: Cloudflare Pages, Netlify, and SFTP.
- Multiple languages: a page per language with links between them.
- Incremental builds for large sites.
- A docs site built with Dapper, with its own
llms.txt, so assistants learn Dapper accurately. dapper doctor: checks the setup (Perl, optional tools like Pagefind and the AWS CLI) and says how to fix what's missing.- Windows support for the single file.
Not planned
- An AI inside Dapper. Dapper stays model-neutral and free; the assistant is the user's own.
- A hosted service. Deploying to the user's own hosting is the point.
- A JavaScript framework or server rendering. Dapper makes plain HTML.
How releases are made
Each release keeps the bar set by 1.0: every feature has tests that fail without
it, the retired markbenson.io site still builds the same, dapper check
passes on every starter, the README is updated and its examples are
built before they're documented, and CI passes on Perl 5.10 to 5.40, the
single file and the Docker image.