Gin for Claude Code
Claude Code marketing you swipe to post
Claude Code marketing with Gin takes one pasted message: a SessionEnd hook sends each session's summary to Gin, and nothing posts until you swipe right.
Sign in at trygin.ai/app, copy the setup prompt with its one-time code, and paste it into Claude Code. It installs Gin's CLI and plugin. The hook sends the session's summary, commit subjects and pull request links, never diffs. Gin picks the channels and writes one post for each, as a card in your deck.
Sample data
Why a session summary makes a good post
At the end of a session, Claude Code usually writes a short account of what it did: the feature, the fix, the tests, the PR it opened. That paragraph is the most honest changelog most teams have. It's written minutes after the work, in plain words, by something that saw every step.
It's also the thing that gets lost. The terminal scrolls, the next task starts, and by Friday nobody remembers what shipped on Tuesday. Marketing a product built with a coding agent is mostly a capture problem: get that summary somewhere before it disappears, then turn the parts a customer would care about into posts.
You can do the capture yourself with a hook (there's a ten-line version further down), or let Gin do the capture and the writing. Either way, the summary is the raw material for the weekly ship note our build in public guide describes.
Setup: paste one message into Claude Code
Sign in with Google at trygin.ai/app. Gin shows a message with a one-time code. Paste it into Claude Code in the repo you want to market:
Set up Gin for this project.
Read and follow https://trygin.ai/setup.md
My one-time code: GIN-XXXX-XXXXThe code works once and expires after 10 minutes. Claude Code reads the setup file and runs:
npx @winding-labs/gin setup --code GIN-XXXX-XXXXThat command does the rest:
- Installs the
ginCLI and the Gin plugin for Claude Code (or for Codex or Cursor, if that's what you're running). - Turns on the session and merged-PR hooks.
- Finds the products in the repo, such as a site or an app.
- Sends a test message, "Gin setup check from Claude Code", which lands in Incoming in the app.
- Prints connect links for LinkedIn and X. You open them and sign in yourself.
What the SessionEnd hook sends, and when
In Claude Code, hooks are commands that run at fixed points in a session's life. Anthropic's docs list over thirty events. Two matter here:
- Stop fires "when Claude finishes responding". That's every turn, many times per session, and a Stop hook can even keep Claude working.
- SessionEnd fires "when a session terminates". It can't block anything; the docs describe it as a place for side effects like logging or cleanup. SessionEnd hooks share a default budget of 1.5 seconds, which a longer per-hook timeout can raise.
Gin uses SessionEnd, so one session becomes at most one message. Its plugin ships a hooks/hooks.json with a single SessionEnd entry that runs gin ship --session - --quiet with a 10-second timeout. Per Anthropic's docs, a plugin's hooks merge with your user and project hooks while the plugin is enabled.
The CLI then reads the session's own transcript and sends three things:
- the final summary, meaning Claude's text after its last tool call;
- the subjects of commits made since the session started that the transcript shows it made;
- links to pull requests it opened.
It never sends a diff, a file's contents, your prompts or tool output. It runs only in repos linked with gin setup or gin products add. Merged pull requests reach Gin as messages too. On Gin's side, secrets are stripped before anything is saved, and message bodies are deleted after 30 days.
A worked example: one session, two drafts
The session below is an illustrative example for Zest, the team's own product (standups for coding agents). It is not a real transcript, and the drafts were written for this page in Gin's format.
Added a 6am standup email for every workspace. Each coding agent gets three lines: shipped, stuck, next. Send time follows the workspace time zone, with tests for daylight saving changes. Opened PR #2203; CI passed and it merged.
Commits: feat(standup): 6am email per workspace · test(standup): time zones and DST · fix(standup): skip agents with no sessions
Your coding agents worked all night. Zest now emails a three-line standup for each one at 6am: shipped, stuck, next. In your time zone, daylight saving included.
In this example, Gin routes the message to X (builders who run several agents) and to the LinkedIn page (team leads who read standups). A separate session that only ran "deps: bump wrangler" gets no channels: internal work.
We kept asking our agents what they did overnight. So we built a standup for it.
Starting today, Zest emails a short standup at 6am in your workspace's time zone. Three lines per coding agent:
• what shipped
• what's stuck
• what's next
If an agent didn't run, it isn't in the email. No dashboard to open.
If your team runs Claude Code, Codex or Cursor overnight, this is the first thing to read in the morning.
Note what made it into the posts and what didn't. The commit about time zones became "in your time zone". The test file names and the PR number stayed out. Every claim traces back to the summary.
Write commits and PR titles that become good posts
Whatever tool turns your work into posts, it can only use what the session recorded. Small habits make the raw material better:
- Ask for a customer-facing summary. Add a line to your CLAUDE.md: "End each session with two sentences on what changed for users." The summary is where a post starts.
- Put the user-visible change in the PR title. "Standups at 6am in every time zone" beats "refactor scheduler".
- Keep internal work recognizable. Conventional prefixes like chore, deps and ci make it easy to skip what nobody outside cares about.
- Mention the number when there is one. "Cuts build time from 9 to 4 minutes" gives a post something concrete.
- Never put secrets or customer names in commit subjects. They travel further than you think.
The CLI and the MCP server
Outside the hook, you or Claude Code can talk to Gin directly. The CLI:
gin status
gin ship "Standups now arrive at 6am in each workspace's time zone"
gin draft "the new 6am standup" --channel linkedingin ship sends a message the same way the hook does. gin draft writes a post for one channel. Setup also gives Claude Code Gin's tools: the plugin brings Gin's MCP server, which runs locally over stdio, so there's no key to paste. To add it by hand after setup:
claude mcp add gin -- ~/.gin/bin/gin mcpOther MCP clients take the same server under mcpServers, the line setup prints for Cursor: "gin": { "command": "~/.gin/bin/gin", "args": ["mcp"] }. Per Anthropic's MCP docs, claude mcp add uses the local scope by default: this project only, stored in ~/.claude.json. Add --scope user to load it in all your projects, or --scope project to write a .mcp.json your team shares. In interactive sessions, Claude Code asks for approval before it uses servers defined in a project's .mcp.json.
Through the MCP tools, Claude Code can send a message about what shipped, draft or remix a post, add a product, or print a connect link for you to open. Our LinkedIn MCP guide covers what any MCP server may do on LinkedIn, official or not.
What never happens without your swipe
Gin gives your coding agent an agent key (it starts with gin_). That key can send messages, draft and manage products. It can never:
- publish a post or swipe right;
- turn on autopilot;
- start or raise a boost;
- touch billing.
Those need a person's session in the app. Connecting LinkedIn or X needs you too: the agent prints a one-time link and you sign in. Every post waits as a card until someone swipes right on it.
After three right swipes in a row on a channel, Gin offers autopilot for that channel. Even then it pings you two hours before each post, and one tap pauses everything. Anthropic's docs make a related point about plugins in general: whatever a plugin runs, it runs as you, so read what it does before you install it.
Gin doesn't reply to people, send DMs or post to Facebook. It writes and publishes posts about what you ship, and only after a swipe.
Codex
OpenAI's Codex has hooks with similar names: SessionStart, SessionEnd, PreToolUse, PostToolUse, Stop and more. The docs say hooks are enabled by default, and that hooks an administrator didn't set up must be reviewed and trusted before they run, so Codex may ask you to approve Gin's hook once. A Stop hook receives the session id, the transcript path and the last assistant message.
Gin's Codex plugin uses Stop, but it doesn't send on every turn. It waits until Codex has been quiet for 10 minutes, then sends what the session shipped, once per session, with the same rules: summary, commit subjects and PR links, never diffs.
For MCP, Codex has its own commands: codex mcp add <name> --url <url> for a remote server, or a [mcp_servers.<name>] table in ~/.codex/config.toml. Gin's Codex plugin bundles its MCP server, so setup covers it.
Cursor
Cursor's hooks live in a hooks.json with "version": 1, either in the project at .cursor/hooks.json or for your user at ~/.cursor/hooks.json. Its stop event is "called when the agent loop ends" and reports whether the loop completed, was aborted or hit an error. Event names are camelCase here: sessionStart, afterFileEdit, stop.
Gin's Cursor plugin hooks into stop and runs the same gin ship command. Cursor reads MCP servers from .cursor/mcp.json in a project or ~/.cursor/mcp.json globally, and supports stdio, SSE and Streamable HTTP servers.
No Gin? Keep a ship log with one hook
If you only want the capture, a SessionEnd hook can append your recent commit subjects to a file in the repo. Add this to .claude/settings.json (or ~/.claude/settings.json for every project):
{
"hooks": {
"SessionEnd": [
{ "hooks": [ { "type": "command", "command": "~/.claude/hooks/shiplog.sh", "timeout": 10 } ] }
]
}
}Every hook gets a JSON object on stdin with fields like session_id, transcript_path and cwd. The script uses jq to read the working directory:
#!/bin/sh
# ~/.claude/hooks/shiplog.sh: append recent commit subjects to SHIPLOG.md
cwd=$(jq -r .cwd)
cd "$cwd" || exit 0
{
echo "## $(date +%F)"
git log --since="6 hours ago" --pretty="- %s"
} >> SHIPLOG.mdMake it executable with chmod +x. At release time, paste the new lines from SHIPLOG.md into the generator on our release notes template page. It drops the internal ones and fills in a changelog entry, customer notes and a post to edit.
Questions
Does Gin read my code?
No. The hook sends the session's final summary, the subjects of commits the session made and links to pull requests it opened. It never sends diffs, file contents, prompts or tool output, and it only runs in repos you linked.
Can Claude Code publish posts through Gin on its own?
No. An agent key can send messages and draft posts, but it can never publish, swipe, turn on autopilot, start a boost or touch billing. A person swipes right in the app for every post.
Which hook does Gin use?
SessionEnd in Claude Code, so each session sends at most one message. In Codex it uses Stop and waits until the session has been quiet for 10 minutes. In Cursor it uses the stop hook.
Does it work in Claude Code on the web?
Anthropic's docs say cloud sessions, including claude.ai/code, do not load plugins from your local settings, so Gin's session hook, installed on your machine, does not run there. You can still send a message with gin ship from a machine where the CLI is installed.
What does it cost?
Free covers up to 3 products and 20 published posts a month, with one seat. Paid plans add posts, seats, autopilot and boosts; trygin.ai/pricing has the prices. Drafts, swipes and remixes are free on every plan.
Related guides
Sources
Checked November 19, 2026. Prices and limits change; each link goes to the page we read.
You ship. Gin posts.
Get the app, paste one message into your coding agent, and swipe through your first cards tonight.