Your coding agent writes the user docs from your codebase, screenshots included.
io.github.SupportPages-io/wtfm MCP Server
io.github.SupportPages-io/wtfm is an MCP server identified by the slug io-github-supportpages-io-wtfm. Its description states that it supports a workflow where a “coding agent writes the user docs from your codebase, screenshots included.”
🛠️ Key Features
Generates user documentation from a codebase
Includes screenshots in the produced docs
🚀 Use Cases
Producing user documentation derived from existing source code
Documenting features/screenshots alongside written content
⚡ Developer Benefits
Connects documentation output directly to codebase content
Bundles screenshots with the generated user docs
⚠️ Limitations
Limited information available beyond the stated description
Claude Code writing an illustrated guide to adding a site in Plausible Analytics: the prompt, the writer's progress, and the finished article with screenshots Real output from plausible/analytics, replayed fast. A real run takes a few minutes.
WTFM is a plugin for your coding agent that writes your user docs. Ask
Claude Code or Codex
for a guide and it reads your source, writes the steps in plain language, and shows every
screen with a screenshot rendered from your real templates and CSS. No running app, no
manual captures, no account.
Not API reference, not a README: docs for the people who will never read your code,
like the customer who can't find the export button, or the admin inviting their team.
WTFM was formerly SupportPages Writer. The supportpages command still works as
an alias for wtfm, and existing installations, settings and SUPPORTPAGES_*
environment variables are unchanged.
Set it up once in your terminal, then just ask your agent:
sh
curl -fsSL https://wtfm.sh/install | bash
sh
wtfm setup # once per computer
sh
wtfm init # once per project
You → Claude Code: Write an illustrated guide to inviting a teammate.
Features
Written from your source. Routes, button labels and flows come from the code,
not from a description you type, so the steps match what users actually see.
Screenshots without running the app. Mockups are rebuilt from your own markup,
styles and branding, then rendered to PNG in a local headless Chromium.
Plain Markdown output. Each article is an index.md with its images beside it,
ready to commit next to your code or drop into any docs site.
Checked before it's saved. Every article passes structure, copy and screenshot
checks first.
Uses the agent you already have. Claude Code or Codex, with your own subscription
and model settings.
Local-first. Nothing is sent to SupportPages.io unless you choose to publish.
Examples
The same loop on other open-source codebases. Each image is one screenshot from a
guide WTFM wrote from that repository's source; the apps were never run.
For a whole help centre, see WTFM's own docs: every
article and screenshot there was written by WTFM from this repository.
Install
sh
curl -fsSL https://wtfm.sh/install | bash
The installer downloads a checksum-verified release with its own Node.js, installs
the wtfm command into ~/.local/bin (no sudo), and offers to add it to your
PATH. It also installs supportpages as an alias for wtfm.
You'll need:
macOS or Linux, on ARM64 or x64 (on Windows, use WSL)
Chromium for screenshots is downloaded the first time you run wtfm setup.
Updating: releases check for updates once a day and install them in the background.
Run wtfm update to update now, or set SUPPORTPAGES_AUTO_UPDATE=0 to turn
automatic updates off.
Install in Codex
WTFM is also a Codex plugin. You'll need Node.js 22.12 or later with npm on your
PATH.
Then start a new Codex session. The first time the plugin starts, it installs its npm
packages (without running their install scripts) and downloads Chromium for
screenshots in the background. Codex asks you to review the plugin's two hooks in
/hooks before they run; WTFM works without them. To update, run
codex plugin add wtfm@supportpages again.
Build from source
You'll need Node.js 22.12 or later, npm and Git.
sh
git clone https://github.com/SupportPages-io/wtfm.git
cd wtfm
./install-cli.sh
This builds your checkout and installs it as wtfm (with the supportpages alias).
Run ./install-cli.sh again after pulling changes.
Quick start
Configure once in the terminal, then do everything else in your coding agent.
1. Set up your computer
sh
wtfm setup
This connects Claude Code and/or Codex, adds the SupportPages.io writer to them and
downloads the screenshot renderer. When it asks how you want to work, choose
Save articles in my projects without an account.
2. Set up your project
sh
cd path/to/your-product
wtfm init
Pick where articles should be saved and which model to use. WTFM then
studies the project's structure, styles and branding once, so every article starts
from that analysis.
3. Ask your coding agent
Open Claude Code or Codex in your project (restart any session that was already open)
and ask for an article in plain words:
Write an illustrated guide to inviting a teammate.
A few minutes later the article is in your project:
Write a troubleshooting article for when a sign-in link has expired.
CLI commands
You only need the terminal to set things up and keep them running. Writing happens in
your coding agent.
Command
What it does
wtfm setup
Set up this computer: connect your coding agents and install the renderer
wtfm init
Set up this project: where articles go, which model, project analysis
wtfm configure
Change this project's agent, model, writing style or article folder
wtfm analyse
Re-run project analysis after big changes (--refresh)
wtfm status
Show this project's setup and article progress
wtfm doctor
Check that everything is installed and connected
wtfm telemetry off
Stop anonymous usage counts and crash reports (on, status)
wtfm update
Install the latest release
wtfm uninit
Forget this project's setup (keeps a recovery archive)
wtfm remove
Disconnect from your coding agents (keeps projects and articles)
wtfm uninstall
Remove WTFM from this computer; --purge-config also signs out and forgets project bindings
Run wtfm --help for every option. status and doctor accept --json.
Your articles
Where they go.output/articles/<slug>/ by default. To keep them in your repo,
choose a folder like docs/help during init or with wtfm configure.
Writing style. Pick from friendly, minimal, technical or formal with
wtfm configure.
Screenshots. Each article gets up to three. The main path gets them first;
optional branches are described in text. Every screenshot carries a small
"created automatically by supportpages.io" strip.
How it works
setup connects a local MCP server and a
writer agent to Claude Code or Codex.
init has your agent map the project once: framework, routes, layouts, CSS and
branding.
When you ask for an article, the writer agent reads the relevant source, writes the
steps, and builds each screenshot as an HTML mockup of your real UI.
The mockups are rendered to PNG with a local Chromium, checked against your source,
and saved with the article.
Publish
Give your users somewhere to read it. A Markdown folder in your repo helps nobody
who can't find it. SupportPages.io,
made by the same team, turns the guides WTFM writes into a help centre your users can
search and ask: on your own domain, in your branding, and kept current as the product
changes.
Kept current on every merge. When a pull request merges, SupportPages.io reads
the diff and drafts the article updates for you to review.
Video walkthroughs. Each guide can come with a narrated video that clicks through
the same steps, with captions.
Answers, not just search. Readers ask in their own words and get an answer drawn
from your articles, with the sources linked.
Your branding on your own domain, an editor to review drafts before anyone sees them,
and publishing straight from your coding agent.
Everything above works without it, and a free account hosts up to 10 articles.
The same pattern applies: connect once in the terminal, then work in your agent.
Connect a project:
sh
wtfm publish
You'll sign in or create an account in your browser and choose a help centre. Any
articles you've already saved upload as drafts. (Setting up a new project? Choose
Publish them to a SupportPages.io help centre during wtfm init instead.)
Then keep asking your agent:
Write a guide to exporting a report.
New articles arrive in the help centre as drafts, with a link to review them in the
editor. Publishing is always a separate step: click Publish in the editor, or ask:
Publish the guide to exporting a report.
Command
What it does
wtfm publish
Sign in, pick a help centre and upload this project's saved articles
wtfm yolo
Write the whole manual on SupportPages.io (account required; drafts only)
wtfm login / logout
Sign this computer in or out
wtfm sync
Refresh help-centre settings and article history
A help centre never needs access to your GitHub or Bitbucket. Only wtfm yolo does.
wtfm yolo: write the whole manual
Starting from nothing? wtfm yolo plans every guide your product needs and
writes them all at once, on SupportPages.io's servers instead of your computer and your
coding agent's usage.
sh
wtfm yolo
It walks you through whatever is missing: browser sign-in, choosing or creating a help
centre, and connecting your repository (SupportPages.io needs read access to write from
your code). It then shows your plan's article allowance and asks before starting
(--yes skips the question). The free plan covers your first 10 guides.
Progress shows live: sections, recommended articles, then drafts as they are written.
Everything lands as drafts; nothing goes live until you publish it in the editor, whose
link is printed at the end. Ctrl+C only stops watching: the run keeps going, and
wtfm status picks it up again. If a run fails, run wtfm yolo again to resume it
without losing the drafts already written.
Privacy
Local projects send only anonymous usage counts and crash reports to
SupportPages.io (see below). Your coding agent and its model provider read your
source, as they do for any other task.
Publishing uploads the article, its images and a small provenance record.
Your source code is never uploaded.
wtfm yolo writes on SupportPages.io from the repository you connect there;
it is the only command that gives SupportPages.io access to your code.
Credentials are stored in private files under ~/.config/supportpages, never in
your repository. Signing in also saves your account email there, to show which
account is connected.
Other network requests carry none of your code or articles. Chromium for
screenshots downloads from Google's Chrome for Testing storage, and the plugin's
packages install from the npm registry. When a project's own styles can't be
compiled, analysis fetches its CSS framework (Bootstrap, Bulma or Foundation) from
the jsDelivr CDN, and a mockup with no project styles loads Tailwind from its CDN
while rendering. An article's optional generated images use OpenAI only if you
have set your own OPENAI_API_KEY; WTFM never sets it.
Project analysis runs unattended.init and analyse start your agent with its
permission prompts off (--dangerously-skip-permissions for Claude Code,
--dangerously-bypass-approvals-and-sandbox for Codex). Writing articles uses the
agent's normal approval prompts.
Plugin hooks
Installed as a Claude Code plugin, WTFM registers two hooks. Neither sends anything
over the network except the Chromium download, and neither can block or change a
tool call.
SessionStart checks whether Chromium for screenshots is installed. If it
isn't, it starts puppeteer's Chromium download in the background, once, and
returns straight away without printing anything into the session.
PostToolUse runs after every tool call, but does nothing unless an article
is being written (a .rtfm-trace/CURRENT marker in your project). During a run it
appends one line per tool call to .rtfm-trace/<article>.ndjson in your project:
the time, tool name, file path, search pattern and the first 500 characters of a
shell command. The trace stays on your computer and is used to spot repeated
work in a run. It prints nothing and always exits successfully.
Anonymous usage counts and crash reports
WTFM tells us how many installs are in use and when something breaks, so we
can fix it. It says so the first time you set up a project, and it's easy to turn
off. It is sent to SupportPages.io without your sign-in and stays anonymous until
you sign in from this computer; after that, usage from this computer is linked to
your account. Turning it off stops both.
What is sent
A random install ID created on this computer (not derived from it), the WTFM
version, the coding client's name (e.g. claude-code), your OS, CPU architecture
and Node.js major version.
A count when WTFM is first used and when it is uninstalled (and whether its
settings were kept), when a project is set up (local or hosted),
and when an article or video walkthrough finishes: its outcome, where it went
(local, uploaded or generated on SupportPages.io), an error code if it failed,
and how many seconds it took; likewise when a wtfm yolo run starts and ends.
For unexpected errors only: the error type and code, and stack frames with every
path outside WTFM replaced by <external>. Messages from other code are
never sent.
What is never sent: your code, file names or paths, article titles or content,
repository or project names, or credentials. Your IP address is not stored. The
install ID is also sent when you sign in from this computer, which is what links
later usage from it to your account.
Turn it off with any of these:
ask your coding agent to turn off SupportPages telemetry;
run wtfm telemetry off;
set SUPPORTPAGES_TELEMETRY=0 or DO_NOT_TRACK=1.
It is always off in CI. wtfm doctor shows whether it's on and why; set
SUPPORTPAGES_TELEMETRY_DEBUG=1 to print each report to stderr as it is sent.
FAQ
Is this for API docs or READMEs?
No. WTFM writes end-user documentation: how-to and troubleshooting guides for the
people who use your product, with screenshots. Your agent already handles developer
docs well.
Why not just ask Claude Code to write the docs?
It will write you decent steps. It won't show the screen, and it guesses at labels it
hasn't looked up. WTFM adds a project analysis, a renderer that builds screenshots from
your own UI code, and checks that run before anything is saved.
What does it cost?
WTFM is free and Apache 2.0. Articles run on your own Claude Code or Codex subscription.
wtfm yolo is the exception: it runs on SupportPages.io and needs an account; the free
plan covers 10 guides.
wtfm: command not found
Open a new terminal, or run the PATH command the installer printed.
My agent can't see the SupportPages tools.
Restart any Claude Code or Codex sessions that were open during wtfm setup.
Setup won't finish.
Run wtfm doctor, fix what it reports, then run wtfm init again.
An article was interrupted halfway.
Ask your agent to continue it, or run wtfm status to see where it stopped.
Does it collect usage data?
Only anonymous counts and crash reports, never your code or articles. See
Anonymous usage counts and crash reports
to see exactly what is sent or turn it off.
Bug reports and pull requests are welcome. When reporting a bug, include your OS,
wtfm --version, the command you ran and its output, with credentials and
private code removed. For larger changes, open an issue first so we can agree on the
approach.
To work on the code:
sh
npm ci
npm test
node scripts/cli.mjs --help
The article engine in engine/ is synced from upstream on each engine release, so
changes there are reviewed here and ported back by the maintainers.
License
Apache 2.0. The UI stylesheets bundled under engine/ keep their own
licenses; see NOTICE. The license doesn't cover the SupportPages name or logo.