Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Jig add the GSAP snippet to my project"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Jig is a private, self-hosted library of code snippets, notes and credentials that your AI coding agents can reach over MCP.
Ask Claude Code, Codex or Kimi to "add the GSAP snippet from Jig" and they search your library, fetch the files, install the dependencies and wire it into the project, following the instructions you saved with it. Every edit is a new version, so when a snippet stops working you can see exactly what changed and roll it back, from the dashboard or from an agent.
Each person runs their own copy, with their own database. Nothing is shared with anyone else.
Step | Example | |
01 | Deploy your copy | One click on Vercel, with a free Neon database created for you. |
02 | Connect your agents | Create a token on the Connect page and paste the command it shows. |
03 | Ask for a snippet | "Add my Lenis setup from Jig": the agent finds it, installs it and adapts it. |
If it speaks MCP over HTTP, it can use your library.
✅ You copy the same setup code between projects: Lenis, GSAP, a WordPress AJAX handler, a Shopify section
✅ You want your AI agents to use your code, not whatever they guess
✅ You want every change versioned, with a diff and a one-click rollback
✅ You keep runbooks, checklists and client notes next to the code
✅ You need somewhere safe for logins and API keys that agents can never read, and notes you can lock behind Touch ID
✅ You'd rather own your data: your server, your database, MIT licensed
Related MCP server: ClawStash
Without Jig | With Jig |
❌ The good version of that script lives in a project from two years ago, and you can't remember which one. | ✅ Search your library by name, tag, language or the code itself. |
❌ Your agent writes its own smooth-scroll setup from scratch, differently every time. | ✅ It fetches your setup, installs the dependencies and follows your notes. |
❌ Someone "improved" a snippet and now it breaks on Safari. | ✅ Diff any two versions and roll back in one click, from the dashboard or an agent. |
❌ API keys sit in a notes app, one paste away from an agent's context. | ✅ Secrets are encrypted in Credentials, which agents can't read at all. |
❌ Sending a client their login means a password in an email. | ✅ A share link that needs a passcode, expires and locks after 5 wrong tries. |
Click Deploy with Vercel. Vercel copies this repository into your GitHub account and creates a free Neon Postgres database for it. The tables are created the first time the app runs.
Choose a password when Vercel asks for ADMIN_PASSWORD. Use a long one, since it protects everything in your library. Set JIG_TIMEZONE to your time zone, such as Europe/London.
Open your new site and log in.
Connect your agents. Open Connect, create a token for each agent and run the setup command it shows.
Optional: turn on credentials. The first time you open Credentials, click Generate a key, add it to Vercel as JIG_ENCRYPTION_KEY and redeploy. Keep a copy in your password manager.
Recommended: run it next to your database. In Vercel, open Settings → Functions → Function Region and pick your Neon database's region (shown under Storage), then redeploy. Every page makes a few queries, and in the same region each one takes milliseconds.
The Connect page shows these commands with your URL and token filled in.
# Claude Code (available in every project)
claude mcp add --transport http --scope user jig https://YOUR_DOMAIN/api/mcp \
--header "Authorization: Bearer jig_..."
# Kimi Code CLI
kimi mcp add --transport http jig https://YOUR_DOMAIN/api/mcp \
--header "Authorization: Bearer jig_..."# Codex: ~/.codex/config.toml, plus export JIG_TOKEN=jig_... in your shell profile
[mcp_servers.jig]
url = "https://YOUR_DOMAIN/api/mcp"
bearer_token_env_var = "JIG_TOKEN"Tokens are stored as SHA-256 hashes, shown once, and can be revoked on the Connect page. Each version records which token saved it.
Tool output is formatted markdown, ready for an agent to read.
Tool | What it does |
| Find by keyword, language or tag. Also searches file contents. |
| Files, dependencies and integration instructions, latest or any version. Suggests matches for a wrong slug. |
| Save a new snippet (one or more files). |
| Save a new version. Only changed fields are needed. |
| History with change notes and who made each change. |
| Unified diff between two versions. |
| Roll back by saving an old version as the newest one. Nothing is ever deleted from history. |
| Find notes by keyword or tag, including the note text. |
| A note's text, latest or any version. |
| Save a note or a new version of one. |
| Note history and rollback. |
Snippets and notes can only be deleted from the dashboard, never by an agent. Credentials are never available over MCP. When filtering by language, javascript also finds TypeScript, JSX and TSX, and css also finds SCSS.
Key | What it does |
| The sections, in the header's order (drag the tabs to reorder) |
| New item in the current section |
| Edit the snippet, note or credential that's open |
| Grid or list layout |
| Move through the list |
| Save, on new and edit pages |
| Cancel, on new and edit pages (twice if you changed something) |
The others are ignored while you type, and on new and edit pages, so they can't throw away unsaved work.
Variable | Required | What it does |
| Yes | The password for your dashboard, at least 12 characters. After 10 wrong attempts from one address, logins from it are paused for 15 minutes, and after 100 from anywhere, password logins pause for everyone (a passkey still works). |
| Set for you | Postgres connection string. The Deploy button's Neon database sets it. Leave it empty locally to use a built-in database stored in |
| For credentials | Encrypts saved secrets: 32 random bytes in base64. The Credentials page generates one for you. If it is lost, saved secrets cannot be recovered. |
| No | Mixed into the key that signs the login cookie, together with a random secret Jig makes for itself in the database. Defaults to |
| No | The site's address, e.g. |
| No | Time zone for dates in the dashboard, e.g. |
Changing a variable in Vercel only takes effect after a redeploy.
Your password is the ADMIN_PASSWORD environment variable, so anyone who can sign in to your Vercel account can recover it:
Open your project on vercel.com and go to Settings → Environment Variables.
Find ADMIN_PASSWORD. Reveal it to see it, or edit it to set a new one.
If you changed it, redeploy (Deployments → ⋯ → Redeploy). New values only apply to new deployments.
Changing it signs out every open session, unless you set SESSION_SECRET. To sign out every browser without changing it, use ⋯ → Sign out everywhere. Your snippets, notes and credentials are not affected; credentials are encrypted with JIG_ENCRYPTION_KEY, not the password.
The Deploy button makes an independent copy, not a fork, so it doesn't update itself. To pull in new versions:
git remote add upstream https://github.com/nboada/jig.git # once
git pull upstream master
git pushVercel redeploys on push. New tables and columns are created automatically on the next request.
bun install
cp .env.example .env.local # set ADMIN_PASSWORD
bun run devWithout DATABASE_URL the app uses PGlite (Postgres in WASM) stored in .data/, so there is nothing else to install. To turn on credentials locally, open Credentials and click Create encryption key: it is written to .env.local for you.
Other scripts: bun run build, bun run start, bun run typecheck, and bun run test (data layer, encryption, sharing, rate limiting and MCP tools against an in-memory PGlite).
OAuth, which the claude.ai web and desktop connectors need. CLI agents work with bearer tokens today.
This server cannot be deployed
Shared memory for all your AI agents, your whole team and every MCP client — save, search, recall.
Governed AI agent skills — one library, distributed to devs and exposed to remote agents over MCP.
A shared team wiki your coding agents read and write — across every repo and every MCP client.
Agent-native notes, tasks, dev-docs, vaults, sync & handoffs. MCP + OpenAPI dual surface.
