Traditional clipboard tools only list what you copied; what the content itself looks like is not their business. Hangli adds three layers — it still holds your seven types of history, but it also reshapes content on the spot: one ⌃⌥⌘W turns Markdown into rich text Word understands, the context menu hands it to AI or exports it to a file, and with MCP on, Claude Code and Cursor can come and read it themselves.
Clipboard managers were nothing new, and they increasingly looked alike. We rebuilt one not because history lists aren't pretty enough, but because nobody ever took over the part that comes after the copy.
A screenshot tool from overseas doesn't nag you to save the file. It hangs the shot on a line across the top of the screen: look up when you need it, let it slide away when you're done. The first time I saw it, one thought landed — isn't the clipboard the same? Mostly temporary, single-use, worth being seen and not worth filing.
So the Clothesline became Hangli's first foundation stone: copy and it hangs, look up and it's there.
Almost everything AI hands back is Markdown. Paste it straight into Word and you get **bold** and | tables | everywhere; switch to “keep text only” and all the layout is gone. The only everyday workaround was to find an online converter, paste the content in, convert it, and copy it back out.
Something that should have been solved inside the clipboard took a full detour.
One idea kept things hanging, the other made them useful. If the content is already on the clipboard, then formatting, translating, summarizing and exporting should happen before it leaves. Following the gaps real work kept opening up, custom prompts arrived, then one-key export to DOCX / PDF / XLSX / Markdown / HTML / SVG / XMind, and share images rendered in ten styles as PNG / JPEG / WebP.
Features kept piling up; the starting point never changed.
For over a decade these tools have done the same three things: list what you copied, let you search it, let you paste it again. What the content becomes after the copy has never been anyone's job — Markdown from AI gets smeared into Word as-is, a table screenshot stays a picture. Hangli fills in the two layers above.
The built-in MCP server is read-only and binds to the loopback address only. Whichever agent you work in can look up what you just copied — no switching back to paste it by hand.
This is the stretch traditional tools never touch: what the content becomes once copied. Hangli rewrites the delivery format on the clipboard itself, or writes it straight to a file and hands it to the next tool.
Everything a traditional clipboard manager does, Hangli does too: seven content types sorted automatically, hanging storage at the top of the screen, groups and pins that stop getting lost, and history that follows you across Macs through iCloud.
Clipboard history hangs on a line: three views, eight themes. Glance at the menu bar to bring it down, then ⌘1–⌘0 to copy the first ten items.
⌃⌥⌘W rewrites the Markdown on your clipboard as rich text Word understands — real headings, real lists, real tables.
Text, links, images, video, PDF, files and audio each get their own type. Turn capture on or off per type and filter history by it.
DOCX, PDF, XLSX, Markdown, HTML, SVG — plus rendered share images, notes written into Obsidian, or Apple Notes.
Straight from the context menu, with editable results you can save as new items. Works with any OpenAI-compatible service; the key stays in your local Keychain.
Copy a screenshot of a table, have it recognized as a real table and export XLSX. Local recognition or Baidu cloud recognition, your choice.
A built-in read-only MCP server bound to 127.0.0.1:39817. Claude Code, Cursor and Codex each connect with a single command.
History snapshots sync through iCloud Drive, and merging never drops a pin or a favorite. A companion iOS reader renders Markdown into long images and PDFs.
Press ⌃⌥Space, or rest the pointer on the menu bar for a moment, and the Clothesline unfolds from the top of the screen. It's a borderless floating panel that follows the pointer's display, never steals focus, and hands focus straight back to the app you came from.
AI answers always come back as Markdown. Paste straight into Word and you get **bold** and | tables |; choose “keep text only” and the formatting is lost. Hangli rewrites the clipboard itself, so pasting works in any rich-text editor.
This week's focus is three things:
| Metric | Target | Actual |
|---|---|---|
| Build pass rate | 100% | 100% |
Click here to see what it does
Word, Excel, Markdown, HTML, PDF and share images all live in one export system: document formats do real file writes, tables land directly as worksheets, and share images render as 3:4 cards.
Laid out by the export template as a standalone OOXML document — no Word or Pandoc needed.
Paginated for you, ready to preview and share from the system.
Markdown tables or recognition results written as worksheets.
Exported as-is, ready for any static site or note vault.
Ships with its own preview window; the code is ready to reuse.
Vector output with scripts and external references sanitized away.
Writes both HTML and TSV representations, so whatever you paste into understands it.
Posts a sticky note that floats above every app, with adjustable color, size and opacity.
440×586 by default — the 3:4 card ratio social platforms like. Ten built-in themes. Cover title, body font, palette, line height, page margins, page-number position and ruled lines are all adjustable; long content paginates automatically, and a single --- forces a page break.
Exporting isn't just stripping Markdown symbols — what the output looks like is a separate question. Hangli pulls that out into a config file, the template: fonts, sizes, line spacing, indentation, heading levels, table styling, margins, headers and footers all live in there.
Hangli
.docx you already have and its fonts, sizes, line spacing and table styling are copied over — no retyping parameters# writes .docx / .rtf / .html / .txt directly
swift run HangliConvert docs/sample.md /tmp/hangli-output
{
"name": "Standard",
"sourceDocument": "Q3-report.docx",
"margins": { "top": 2.54, "left": 3.18 },
"body": {
"chineseFont": "PingFang SC",
"size": 12, "lineSpacing": 1.5
},
"table": {
"kind": "bordered",
"cellPadding": 8, "autoHeader": false
}
}
What you copied was a table, so what you export should be a table. Hangli writes the .xlsx itself — no CSV detour, no Excel required.
Markdown tables are written out in order as “Table 1”, “Table 2”…, header row intact and ready to use.
Paragraphs and lists go into a “Body” sheet, one row per paragraph, still editable back in Excel.
A table copied from a web page (HTML or tab-separated text) becomes a single “Table” sheet — no manual realigning first.
Table screenshots and PDFs go through OCR and export to XLSX the same way, so nothing has to be retyped.
Copying a table writes both HTML and TSV, so it pastes as a table into Excel, Sheets or Notion.
Border weight, color and header shading come from the shared settings; the “Body” sheet can get borders too.
Tables from AI are Markdown too. Paste one straight into Excel and the whole thing crams into a single cell, pipes and dash lines included, and you're left doing “Text to Columns” by hand. A screenshot of a table is worse still — you type it out. Hangli parses the header and every row first, then writes the .xlsx itself: open it and it's a real worksheet you can sort and calculate in.
Header, numbers and pipes all sit in A1; Excel has no idea this was four columns by three rows.
Header is a header, numbers are numbers — sorting, filtering and formulas all work.
.xlsx. Recognition results live in memory only and clear on restart.
.xmindAsk AI to lay out a structure and what comes back is Markdown full of hash marks. Turning that into a real mind map normally means: save it as .md, open XMind, create a blank map → import Markdown → save as .xmind. Three or four steps there and back, plus a temp file to clean up afterwards. Hangli writes that outline straight to .xmind — double-click and it is a map you can fold and unfold.
The shallowest heading becomes the root and every deeper level nests one step; lists, tables and code blocks hang off whichever branch is open where they appear.
XMind's own ZIP package: content.json + metadata.json + manifest.json, with the core version declared, so opening it never prompts you to upgrade an old document.
Topic titles longer than 180 characters get cut and tables contribute at most their first 60 rows — the map never turns into a wall of text.
A Markdown outline reads as indentation but it is not a map: you count hash marks to tell levels apart, and collapsing, dragging branches or recolouring them is simply not something it does. Hangli builds the outline into a tree by heading level and writes it in XMind's own package format — the exported file opens straight into XMind, with no import step in between.
The hierarchy is there, but it is only text: to get a map you still save a file, open an app, import, then save again.
The root topic, the branch colours and the one-sided layout are all written into the file — it opens ready to expand and collapse.
AI is optional and off by default. Once enabled, requests go only to the service address you fill in — Hangli has no server of its own.
Right-click text on the Clothesline or in the history list; each action has its own global shortcut.
Summarize, translate and polish are generic actions, but what you actually ask again and again is usually the same few things: pull out the key points, list the action items, rewrite it in your weekly-report voice. Hangli lets you save that sentence as a prompt — give it a name, pick an icon, write the body, then bind a global shortcut to it.
Using AI in a chat box means repeating every step: paste the text, type the prompt, wait, copy the answer back. The prompt itself is something you maintain — write “list it as bullets” today, phrase it differently tomorrow, and the output changes. Hangli turns it into a record: kept on this Mac, shown in the context menu, one click on the selected text and it runs.
You paste the text once and retype this once as well; switch windows and the wording you used last time is gone.
Once enabled it shows up right in the context menu for clipboard text; the ones you are not using can be switched off.
Hangli ships a read-only MCP server that listens on 127.0.0.1:39817 by default (port configurable) and binds to the loopback address only. Every connected tool gets its own token, and Host and Origin are validated against DNS rebinding.
Note that it is not in the context menu: MCP is a read-only door for AI tools. Flip the switch in Settings → AI, then use “Connect AI tool…” to hook up Claude Code, Cursor and Codex one by one.
list_clips — lists items, filterable by type, favorite and pinned, 20 by default and up to 100search_clips — searches by text, title or source appget_clip — fetches the full content by idChoose “Connect AI tool…” in Settings → AI and Hangli writes the config straight into that tool and stores this separate token — no hand-copying JSON.
After changing the port, connected tools need to be disconnected and reconnected.
{
"mcpServers": {
"hangli": {
"url": "http://127.0.0.1:39817/mcp",
"headers": { "Authorization": "Bearer <a token created for this tool>" }
}
}
}
Hangli must stay open while the server runs. Writes, deletions and image content are all unavailable.
These are the defaults. Change them in Settings → Shortcuts, and give each custom prompt a shortcut of its own.
| Action | About | Default shortcut |
|---|---|---|
| Convert to Word | Rewrite the Markdown on the clipboard as RTF rich text | ⌃⌥⌘W |
| Show / hide the Clothesline | Or rest the pointer on the menu bar icon | ⌃⌥Space |
| Copy as plain text | Strip all formatting before it hits the clipboard | ⌃⌥⌘V |
| Table conversion | Turn a Markdown table into a pasteable table | ⌃⌥⌘E |
| Table photo to Excel | Recognize the table screenshot on the clipboard and export XLSX | ⌃⌥⌘I |
| AI Summarize | Summarize the current clipboard text | ⌃⌥⌘S |
| AI Translate | Translate into the target language, direction detected automatically by default | ⌃⌥⌘T |
| AI Polish | Rewrite the wording while keeping the meaning | ⌃⌥⌘P |
Clean a link, strip the formatting, split a sentence, convert to SVG — none of it is big, yet each one normally means hunting for another small app, website or utility. Here you select the content, click once, and it's done.
on* handlers are stripped; Markdown is rendered into a vector image instead.
Network requests happen in exactly three cases, all of them triggered by you:
Everything else runs locally. There is no analytics code in the app.