A Second Brain for Your Coding Agents
Every coding session with an AI tool starts the same way. You explain your architecture. You paste in the relevant files. You remind it about that decision you made last week, the one where you chose Drizzle over Prisma because of the edge runtime constraint.
Then you close the session. And the next time, you do it all over again.
The AI is stateless. Your thinking isn't.
The amnesia problem
If you've used Claude Code, Cursor, or Copilot for anything beyond a one-off script, you've felt this. The tool is good at writing code but it has no memory of why you're building what you're building. It doesn't know your product specs. It hasn't read your architecture notes. It can't recall the thread where you worked through three approaches to auth before landing on the one you chose.
Some people work around this with CLAUDE.md files or long READMEs. Those are great for what they do: describing your codebase, your conventions, your tech stack. AI agents already understand code well. That's not the gap.
The gap is everything above the code. How a feature is supposed to work. Why you chose this approach over three alternatives. What the product requirements actually are. That kind of context lives in your head, or scattered across docs and chat histories, and your AI agent can't access any of it. So it codes in the dark, and you spend half your time turning the lights on.
What if your AI could read your mind?
Not literally. But close.
I've been building Scribbles as a place to think. You write in threads, like texting yourself, and over time your ideas naturally cluster around themes and decisions. It's not documentation you sit down to write. It's context that builds up as you think out loud. And I realized that if my thinking already lives in Scribbles, I should just... give my coding agent access to it.
So I built that. One command:
claude mcp add --transport http scribbles https://api.usescribbles.com/mcp
That's it. No npm package to install. No API key to generate and paste into a config file. The first time Claude Code reaches for your Scribbles data, your browser opens, you log in, approve the connection, and you're done.
And now your AI coding agent has a second brain.
It can search your notes when it needs to understand how a feature is supposed to work. It can read the document where you spec'd out the user flow. It can follow a thread of reasoning, not just the conclusion, but the messy process that got you there. And when it makes a decision during a session, it can write that back as a scribble so you don't lose it.
Instead of spending the first ten minutes of every session re-establishing context, you just start working. The AI already knows what you know, because it can look it up.
And because Scribbles supports teams, this isn't just your context. It's your team's shared understanding. Product specs, design decisions, feature requirements, all accessible to any agent any team member is using.
Your thinking shouldn't be locked in one tool
Your thinking should be accessible from wherever you're building. MCP (Model Context Protocol) makes this possible. It's an open standard that lets any AI tool connect to external data sources.
Scribbles speaks MCP. Claude Code, Cursor, Windsurf, Codex — they all support it. Same server, same access to your entire knowledge base. Your context travels with you.
I think of this as the context layer. A persistent home for the reasoning behind your code that any AI can access.
Try it
If you're tired of re-explaining context every session, connect Scribbles to your agent. Here's Claude Code as an example:
claude mcp add --transport http scribbles https://api.usescribbles.com/mcp
It works the same way with Cursor, Windsurf, Codex, or any tool that supports MCP — just point it at https://api.usescribbles.com/mcp.
And if you don't have a Scribbles account yet — sign up for free and try it today.