I see you

Your AI finally
understands your projects

You renamed a feature in the codebase. But when you switch to marketing, docs, or content — your AI has no idea. It starts blank. Every time. iseeyou.sh gives every project a shared memory.

~/startup/marketing
# One-time setup registers MCP server globally
$ iseeyou-sh setup
Database ready
MCP server registered with Claude Code (claude mcp add --scope user)
 
# Then init each project
$ cd ~/startup/marketing
$ iseeyou-sh init
Parent: Main Product | Permissions: codebase:read domain:read decisions:read
Marketing inherits Main Product knowledge

Claude Code does more than write code

You're a founder. You use AI across everything — building the product, writing marketing copy, creating content, drafting support docs, researching competitors. Each of these is a different project, a different Claude Code session.

But here's the thing: they're all connected. When you change the pricing model in the codebase, the marketing site needs to know. When you add a new feature, the content team needs to create posts about it. When you make an architectural decision, the docs project needs to reflect it.

Right now, none of them talk to each other. Every session starts blank. You re-explain your product to the marketing AI. You re-describe your stack to the docs AI. You repeat the same decisions to every project, every time.

iseeyou.sh connects them. Log a decision in your codebase, and your marketing project sees it instantly. Change your product direction, and every downstream project knows. No re-explaining. No copy-pasting context. Just one shared understanding that flows from parent to child.

You log a decision in the main product manual or auto

"Switched from Stripe to Lemon Squeezy for payments"

Marketing project has access instantly inherited

When you write pricing copy and ask "what payment provider do we use?" — Claude knows. No need to go check the codebase.

Content project works from the same truth inherited

When Claude helps with social posts or blog articles, it has the real product context — not whatever you remembered to tell it

Support docs built on what's real inherited

Help articles, FAQ, onboarding guides — Claude has the actual product state to work from, not a stale mental model

An MCP server. Registered once. Works everywhere.

iseeyou.sh is a Model Context Protocol (MCP) server that registers globally with Claude Code. Once set up, Claude has tools to read and write project knowledge — in every session, every project. Hooks inject context automatically before each prompt.

01

Setup & register

Run iseeyou-sh setup once. It starts Supabase, registers the MCP server globally with Claude Code via claude mcp add --scope user, and saves credentials.

02

Initialize projects

Run iseeyou-sh init in each project. Pick a parent. Set permissions. It scans your files and builds understanding. Claude logs decisions as you work — they propagate to every connected project.

03

Claude just knows

The MCP tools are available everywhere. Context is injected before every prompt. The marketing AI knows the product. The content AI knows the strategy. No re-explaining. Ever.

~/startup/marketing
$ iseeyou-sh init
Workspace: My Startup
Parent: Main Product
Permissions: codebase:read domain:read decisions:read
Found 8 .md files — scanning context...
Marketing inherits Main Product knowledge
Hooks configured. Restart to activate.

# Marketing AI now knows what the product does,
# what decisions were made, and what changed.
# Every session. Without you explaining.

What you need

iseeyou.sh is a self-hosted tool. You run the database, you own the data. Here's how to set it up.

Prerequisites

1.
Node.js 20+
Required to run the CLI and MCP server
2.
Docker
For running local Supabase (Postgres database)
3.
Supabase CLI
Install with npm install -g supabase
4.
Claude Code
MCP server registers globally via claude mcp add
setup
# 1. Install and set up (one-time)
$ npm install -g iseeyou-sh
$ iseeyou-sh setup
Starting Supabase, running migrations...
Database ready, credentials saved
Registering MCP server with Claude Code...
Registered via claude mcp add --scope user
8 tools available in every Claude Code session
 
# 2. Initialize any project (repeat per project)
$ cd ~/your-project
$ iseeyou-sh init
Scanning files, setting up permissions...
Hooks configured. Restart to activate.
 
# 3. Back up your data (run periodically)
$ iseeyou-sh backup
42 rows saved to supabase/seed.sql

Full setup guide on GitHub

Two audiences. One problem.

🚀

For Founders

You use Claude Code for everything — not just coding. Product research, marketing copy, pitch decks, content strategy, support docs. But each project is an island.

  • Marketing AI knows what the product actually does
  • Content is based on real decisions, not memory
  • Stop re-explaining your product to every session
  • Log a decision once, it flows everywhere
  • Dashboard to see the full picture across projects
💻

For Developers

You work across related codebases — main app, admin panel, mobile app, shared libraries. Claude Code doesn't know they share an API or follow the same conventions.

  • Admin panel has access to the main product's documented knowledge
  • Shared conventions across all codebases
  • Logged decisions visible to child projects that have access
  • Permission control — read-only access for some projects
  • MCP tools for Claude to read/write project knowledge

What's under the hood

🗂

Project hierarchy

Parent-child projects up to 3 levels deep. Main product → Marketing → Social media. Knowledge flows down with permission control.

🔒

Permission control

Five categories: codebase, domain, decisions, conventions, task state. Each set to read, write, or none per project. A child can never exceed its parent.

Global MCP server

Registered once with claude mcp add --scope user. 8 tools available in every Claude Code session. Hooks also inject context before every prompt automatically.

📜

Decision ledger

Append-only log with rationale, alternatives, and supersedes tracking. Log manually or let Claude do it. Decisions propagate to child projects.

📋

Task state

What's in progress, blocked, completed, next. Parsed from your TODO.md on init, updated by Claude as you work.

📁

Context from existing files

Init scans your CLAUDE.md, README, session notes, docs — everything. No manual data entry. Your project model is built in seconds.

🌐

Dashboard

Visual project tree, decision browser, task state viewer. See everything across all your projects at a glance.

🔧

Open source & self-hosted

Your data stays on your machine. Supabase backend, MIT license. No vendor lock-in. No cloud dependency.

Stop re-explaining.
Start connecting.

One command to give your AI a memory that spans every project.

$ npm install -g iseeyou-sh && iseeyou-sh setup


Star on GitHub