Fixes
Templates that used elseWhen() after when() showed every branch instead of one. They now show only the branch that matches, so render those templates again.
Notable changes to the makesPDF API, catalog, and product. Newest first.
Templates that used elseWhen() after when() showed every branch instead of one. They now show only the branch that matches, so render those templates again.
Make's MCP Client connects to the makesPDF MCP server with an API key, so Make scenarios and AI Agents can render markdown and templates to PDF and download the file.
The makesPDF MCP server now has setup guides for n8n, Zapier, Activepieces, Windmill, Tines and Workato; each connects with an API key.
A cell in an HTML table that sets no valign is now centred vertically, the way a browser and GitHub draw it. Previously it sat at the top of its row, which stood out on cells that span several rows.
Creating a makesPDF account from inside an AI assistant's connector prompt used to strand you after verifying your email — the assistant waited for a sign-in that never arrived. Verifying now returns you to the assistant and finishes the connection.
Non-Latin document properties, CJK table columns, headings stranded at the foot of a page, footer page numbers in templates, inline code in table cells, frontmatter margins and list keywords, and cramped diagram shapes are all fixed. Export your documents again to pick them up.
makesPDF now installs as an Antigravity plugin and a Gemini CLI extension with one command. Sign in with your makesPDF account in the browser and the agent can validate, preview and render PDFs, with no API key to copy.
Raw HTML table blocks in markdown now render as real tables — merged cells with colspan and rowspan, header cells, alignment and cell backgrounds. Unsupported table HTML falls back to readable text with a warning.
A table cell can now cover several rows with attr.rowspan — the surrounding rows make room for it, stay together across a page break, and the tagged PDF carries /RowSpan and /ColSpan.
Zoroastrian and Middle Persian texts now render with no font to upload — Noto Sans Avestan and Noto Sans Inscriptional Pahlavi join the built-in fallback, drawn in correct right-to-left order and extracted in logical order.
New integration guides and packages. Render every page of an MkDocs, Docusaurus or Hugo site to PDF at build time, or give a LangChain, LlamaIndex, CrewAI, LangGraph or Vercel AI SDK agent a tool that makes PDFs.
render_markdown and render_template now return a short-lived download URL an agent can curl to disk, plus a get_render_download_url tool that re-issues one for any render you own.
Markdown notes that use <sup> — Wikipedia-derived transliterations, ordinals, footnote-style markers — now draw the run smaller and raised on the same line, with no gaps around it.
Hebrew text now renders in correct right-to-left order with no font to upload — including mixed Hebrew/Latin lines, numbers, mirrored brackets, and right-aligned paragraphs. Copy-paste returns your source text, not the reversed glyphs.
Zero-width spaces, bidi marks and other invisible characters that arrive with copy-pasted text — including soft hyphens and byte-order marks — now draw nothing, so a document no longer picks up a stray kanji, an extra space or an unexpected hyphen in the PDF.
Starting a connection from Claude Code and then signing up or signing in failed with "Registration failed". The flow now completes, and an expired link says so instead of showing a generic error.
claude.ai, Claude Desktop connectors and Claude Code can now add makesPDF by URL and sign in with OAuth. Add the connector, approve access once, and the PDF tools work — no API key, no Request-headers beta.
POST /api/v1/device/token now refuses more than 7,200 polls per hour from a single IP address. A normal `makespdf login` polls one code 12 times a minute, far under the ceiling, so nothing changes for it.
A request carrying an API key we no longer accept now returns 401 with code invalid-api-key. It used to come back as a payment challenge, which told people with a stale or mistyped key that their document needed paying for.
A kanban card's priority badge sat on the card's top border, so the outline looked broken around it. The badge is now inset inside the card's top-right corner.
When a markdown list item starts with a code block, heading, quote or table, the marker now sits on that first line instead of alone above it. A list nested inside a quote in an item also keeps its indentation.
In a Mermaid gitGraph, branch names on neighbouring lanes no longer run together — each label now runs along its own lane, reading bottom-to-top.
In a Mermaid quadrant chart, a point near the right edge no longer pushes its name outside the chart — labels flip to the other side of the dot, or move inwards, so they stay readable.
A `<<fork>>` and a `<<join>>` used to draw identical bars with no label, so you had to read the transitions to tell them apart. Each now carries its state's name, as does a `<<choice>>` diamond.
A label like "Submit" or "Request Changes" used to float beside its transition arrow. It is now centred on the line, over a page-colour gap, the same way flowchart edge labels are.
A radar-beta axis label longer than about ten characters was clipped at the left or right edge of the figure, so "Delivery pace" printed as "elivery pace". The chart now widens to fit its labels.
A canvas whose cards would take seconds to lay out now gets an immediate 400 `canvas-too-complex` saying what to change — split one huge note, or use fewer cards — instead of holding the request. Shrink-to-fit is much cheaper and long words break with a search, so ordinary boards render in the same time or less.
An enhanced PDF whose upload or queue hand-off failed could leave a job stuck at "queued" that a retry polled forever. API-key and signed-in submissions now report the failure straight away, wallet-paid jobs are swept to failed by the daily watchdog, and a cancelled job stays cancelled.
If someone pre-registered your email — with a password, or through a provider that never verified it — signing in with Google or GitHub on that verified address now gets you in, links the provider, and shuts down everything they set up.
A list with a long lazy continuation, or a long run of setext headings, could keep a markdown render busy for tens of seconds before it produced anything. Both render normally now, and a document with more than 150,000 lines is refused up front with a clear error instead of stalling.
Setting `title:` in a markdown file's YAML frontmatter did not change the PDF's title; viewers showed "Document" instead. The frontmatter title is now written to the PDF, through the API, the CLI and the MCP render_markdown tool.
Templates an AI assistant saved or edited through the makesPDF MCP server had an empty History tab. Saves and DSL changes made through MCP are now listed there, the same as changes made in the web app or through the API.
Pass your invoice XML in options.facturX and /render returns a Factur-X / ZUGFeRD hybrid PDF (PDF/A-3A) that is also tagged for accessibility. /preview checks the XML for free.
Local images named like "Photo (2024).jpg" could go missing from PDFs exported with the Obsidian plugin or the VS Code extension. Update and export again.
A table whose header or cell content repeated .table-header or .table-cell produced a PDF that failed PDF/UA-1 validation. Those PDFs now pass, so render the template again.
Template text styled with a numeric weight such as font-weight 600 or 700 was coming out in the regular face. It now renders bold, so run your template again.
The template library at /templates is now public. See each template's rendered page beside its full DSL source and sample data, and download the PDF, with no sign-in.
GFM alerts now show an icon for their type, and a title written after the marker, as Obsidian callouts use, appears in the header instead of on a line below it.
A small set of markdown shapes, such as thousands of unclosed links or emphasis markers in one paragraph, could make a conversion run for minutes. They now return a 400 with code markdown-too-complex and a hint saying what to change.
A line holding only <br> (or another HTML tag) could hide everything below it up to the next blank line. That text now renders, line breaks inside list items are kept, and a numbered list that continues after a bullet list keeps counting.
Code blocks, quotes, tables, headings and extra paragraphs inside a markdown list item used to be missing from the PDF. They now render, in the order you wrote them, inside the item.
Mermaid class diagrams draw the relation label on the line and both multiplicity labels, one at each end — syntax that used to be silently dropped.
Mermaid state diagrams can now contain a diagram of their own, and notes are drawn beside the state they annotate instead of being clipped away.
NotoSans templates finally have a real italic face — `font-style: italic` used to render upright, and every NotoSans bold run reported a missing-font warning. Both are fixed, and existing documents keep their layout.
A single markdown block past ~8MB used to return a bare 500. It now returns 413 with code markdown-block-too-large and advice that fits the block's shape — and the limit is documented.
The per-caller rate limit on /api/v1/md, /api/v1/canvas and /api/v1/render goes from 2,000 to 10,000 renders per hour, because the bookkeeping behind each render no longer runs on the request.
Canvas card content uses an Obsidian theme by default — real Obsidian type sizes and spacing, a node label above every file card, and more cards fit without a continuation page. Pass theme "github" or "default" for the old look.
PDF file nodes on a JSON Canvas board now render as the PDF's actual page — vector, tagged, with #page=N honoured — instead of a placeholder card.
Chinese, Japanese and Korean text now renders out of the box — markdown, templates and canvas boards all draw CJK with a built-in Noto Sans fallback, and the new lang option picks the script and sets the PDF's /Lang.
Every emoji in your PDFs — faces, flags, ZWJ families, skin tones, keycaps — now renders in full colour as native vector graphics, in markdown, templates and canvas boards alike.
A stray check mark, bullet or arrow in your text no longer renders blank. Characters your chosen font lacks are drawn automatically from the embedded symbols font, at the same size and colour.
Send a JSON Canvas document (an Obsidian .canvas board) and get a tagged, accessible PDF back — cards, connector arrows, groups and images, with overflowing cards continuing on later pages instead of clipping.
Markdown to PDF renders can now carry footer page numbers — three formats, three positions, right-aligned by default.
Settings → Account now lists your sign-in methods and lets you attach or detach Google and GitHub — even when the provider's email differs from the one you signed up with.
Markdown documents can now embed Graphviz DOT diagrams — plus a declarative YAML graph block for data-shaped diagrams — rendered as crisp native PDF vector graphics like every other diagram type.
accTitle and accDescr in a Mermaid block now become the figure's alt text in the tagged PDF — a screen reader gets a real description instead of a generic label.
chart() now draws scatter plots with type "scatter" — a point per row on two numeric value axes, with optional x-min/x-max bounds and generated alt text.
chart() can now draw bars and a line in the same chart, with an optional right-hand axis for the rate that needs its own scale — revenue bars plus a margin-percent line.
chart() gains horizontal: true, drawing bars left-to-right with categories down the left axis — the form to use for ranked lists and long category names.
chart() gains stacked: true to stack series within each category, with negative values stacking below zero, an axis fitted to the stack totals, and per-category totals in the generated alt text.
Decision branches now leave a flowchart's diamond from separate points, merge nodes sit centred between their parents, and every Mermaid diagram is centred in the page column.
A base64-inlined PNG used to eat the whole 200KB markdown budget. Inline images are now excluded from the text cap and bounded by the same rules as remote images — 5MB each, 20 per document — with a 15MB request body cap.
A monthly KPI report, a SaaS usage statement and a certificate with a vector seal join the template library — the first library templates to use chart() and vector().
New chart() builder draws bar, line and pie charts from your render data as native PDF vector graphics, with alt text generated from the values.
New Mermaid diagram types you can embed in Markdown → PDF.
Wallet-paid /enhance calls now validate the document before the USDC payment settles — already-compliant, unenhanceable, and previously-failed PDFs get a free 400 instead of a paid failure.
Claim the wallets you pay with via a personal_sign signature and their USDC renders show up alongside your credit renders in usage history.
A new migration converter takes pasted react-pdf JSX and returns validated makesPDF builder DSL with a live PDF preview.
Starter now has an annual option - $100 a year, the price of 10 months, with the same 1,500 credits every month.
Organisations can now prepay makesPDF credits by invoice with a PO number, without talking to anyone.
A table row, blockquote, code block or alert taller than a single page used to fail the whole render with a 500. It now splits across pages, keeping column alignment and borders intact.
Agents connected over MCP can now render markdown to PDF directly, instead of validating it in MCP and dropping out to curl to produce the file.
Every rendering endpoint now accepts author, subject and keywords, and writes them to both the PDF Info dictionary and the XMP packet. Accessibility checkers were flagging their absence.
The Markdown to PDF page on the site now works signed out, on the same free path the API and the editor extensions have used since May. Paste, convert, done.
Export any Obsidian note to an accessible, archival-quality PDF from the ribbon icon or the command palette. Available now in the Obsidian community plugin directory, no account required.
If an x402 wallet payment settles and we fail to deliver your PDF, the full USDC amount is now refunded to your wallet automatically. Validation errors are caught before payment, so a bad request never charges you at all.
Save a card, set a low-water threshold and a refill amount, and let makesPDF top up your credits automatically.
Markdown documents can now embed Mermaid xychart-beta diagrams — line and bar charts with titled axes — rendered as crisp native PDF vector graphics like every other diagram type.
A new endpoint that walks an existing PDF's StructTreeRoot and gives you back the full tag tree — as JSON for agents and tooling, or as a human-readable outline PDF. Every enhance run now leaves one too.
Pay for a PDF/UA-1 + PDF/A-2A remediation job with a USDC wallet — no API key, no account. The 202 envelope hands you an opaque pollToken to poll the job and download the artifact when it's ready.
POST /api/v1/md now accepts unauthenticated requests for short documents — render Markdown to PDF without signing up. Built for the VS Code plugin funnel, open to any caller.
New barcode() builder generates scannable Code 39 and Code 128 barcodes as native PDF vector graphics — no font, no rasterisation, dynamic values via {{template}} placeholders.
New qr() builder generates scannable QR codes — versions 1–40, all four ECC levels, auto-mode selection, dynamic values via {{template}} placeholders, no font, no rasterisation.
Every time the assistant changes the DSL, the template is saved (or updated) automatically with a one-sentence description of what changed. Each save is recorded as its own revision, and the new per-template page has a History tab that lists them all in order.
Render `templateId: invoice-classic` directly through /api/v1/render — no more fork-then-render. Library and saved templates share one endpoint family.
The Use Template button on /templates is now Modify Template — it drops you into /ai/chat with the template already loaded, rendered, and ready to customize.
/enhance lifts the in-place-only restriction. mode=regenerate rebuilds a clean DSL source you can re-render forever; mode=both ships the tagged original and the regenerated source side-by-side, billed once.
/enhance now runs your PDF through veraPDF before queueing, and refuses upfront when the failing rules require regenerate. /validate surfaces the same recommendation so you can preview the routing decision before spending credits.
Submitting a PDF to /api/v1/enhance now returns 202 with a jobId in under a second. Poll a status URL until the job is ready and we email you when it's done — no more multi-minute HTTP connections that drop on flaky networks.
/api/v1/enhance now bills 10 credits per page on success (up from 1). Each page drives a Sonnet vision call, and the previous rate didn't cover the cost on plan-credit callers. The /enhance web UI shows the new cost before you submit. x402 wallet callers are unchanged at $0.20 / page in USDC.
A new options.acceptRasterFallback flag lets /api/v1/enhance keep going when individual pages can't be reconstructed — failed pages embed as tagged Figure rasters of the original instead of flipping the call to a free partial preview. The output is labelled hybrid and bills at the full per-page rate.
Wallet callers can now pay for /api/v1/enhance in USDC. The 402 challenge declares a per-page price ($0.20 / page) computed from the uploaded PDF, so the amount quoted is always the amount settled.
Drop in any PDF and run the ISO reference validator for PDF/A and PDF/UA. Free, no sign-in, nothing kept. Powered by veraPDF 1.28.2 in a Cloudflare Container sidecar.
/api/v1/enhance now bills 1 credit per page on success. The /enhance web UI shows the cost before you submit and links straight to the credit purchase page when your balance is short.
The MCP server now exposes a billed render_template tool, so agents can finish the loop and produce a real PDF without dropping out to curl or the CLI.
New /compare section with side-by-side breakdowns of makesPDF against the major PDF generation tools. First live page is react-pdf; nine more competitors are in the pipeline.
Templates can ship real fillable AcroForm widgets — text fields, checkboxes, dropdowns — that stay interactive in Reader and Preview and remain tagged for PDF/UA-1 screen readers. Now documented across the docs, skill file, and homepage.
A new /settings/renders page lists every PDF you've rendered in the last 7 days with a one-click download — and the same list is available to API callers at GET /api/v1/artifacts.
Agents and scripts can now hit /api/v1/md and /api/v1/render with a signed USDC payment instead of an account. $0.01 per markdown render, $0.02 per DSL render, settled on Base.
When the AI chat renders a preview against pasted JSON, the workspace pane now shows a copy-pasteable two-step API recipe and one-click Save / Try buttons.
attr.grid now accepts "auto" and "<n>fr" so columns can size to their data instead of a defensive percentage. Header, body and totals rows of one table coordinate automatically.
A dedicated pdf-fonts skill documents per-request and account-level font registration, including the BYO path for non-Latin European scripts. Variable fonts now fail fast with a clear error.
Omit widths in table() and the builder picks for you — every column sizes to content, and the column with the widest measured content automatically absorbs leftover space.
The /ai page is now a sortable leaderboard across five scenarios and 16+ providers, with per-row download links to every generated PDF.
Upload TTF or OTF fonts once at the account level and reference them by family name from any template — no per-request `fonts:` payload required.
Top up between $20 and $500 in a single Growth purchase, and admins can now grant wallet credits directly to any user.
blog.makespdf.com is live, served from the same Worker. Under the hood, the Mermaid pipeline was rewritten as @pdf/mermaid-render — 80% smaller SVG, no fake DOM, Workers-native.
The curated template library (invoices, receipts, quotes, resumes, …) is now browsable over the API. Discover a template, fetch its DSL, fork into a user template, render.
A first-party Model Context Protocol server at /api/v1/mcp lets agents drive the full authoring loop — preview → save → render — without scraping HTML.
Every render is now captured as an artifact (inputs + output) addressable by an X-Artifact-Id header, with user and admin pages to file and triage rendering issues.
The biggest pivot since launch. /api/v1/render now takes a templateId you saved earlier, /preview is the free authoring surface, and AI is no longer part of the core worker.