io.br8n/mcp — MCP Server for Portable Working Memory

The io.br8n/mcp server is a small Model Context Protocol (MCP) server that lets a model read an “owned, portable working memory.” It serves a plain-files Markdown “brain template” stored in files you own, described as enabling models to cite that content directly rather than rely on another system’s login.

🛠️ Key Features

Owned, portable working memory
Plain-files Markdown “brain” template
Tiny MCP server for reading the files by any model
Topics: ai-memory, knowledge-base, markdown, mcp, model-context-protocol, portable-memory

🚀 Use Cases

Teams that want chat tools to use knowledge from files they own
Switching tools without losing access to the same working memory
Building model-accessible knowledge bases via Markdown documents

⚡ Developer Benefits

Simple file layout and an “open part” for portability
Model context sourced from readable, citable files

⚠️ Limitations

The open repo excerpt does not describe retrieval, governance, or hosted capabilities; it only notes that these are in a “hosted install.”

br8n

Owned, portable working memory. A plain-files brain template and a tiny MCP server that lets any model read it.

Everything your team has taught a chat tool about how the work gets done is sitting inside somebody else's login. Switch tools and you're a stranger again. This repo is the other direction: the memory lives in files you own, and the model is just a reader.

br8n is the AI delivery practice of Branded Mayhem Collective. This is the open part: the file layout and the door. The hosted install adds retrieval, governance, and someone running it with you — on the same files, which never change shape. br8n.io

What's here

template/brain/ — the brain layout: how-we-work/, decisions/, exceptions/, handoffs/, voice/. Markdown only. One file, one thing. Write the why, so the model can push back later.
src/ — an MCP server (stdio) with three tools: brain_list, brain_read, brain_search. Search returns file + line so answers cite their source. No vectors, no index, no account. Grep is the point.

Use it

bash

npx @br8n/mcp ~/my-brain                         # MCP server on stdio (or: npm i -g @br8n/mcp && br8n ~/my-brain)
# template: git clone https://github.com/Branded-Mayhem-Collective-LLC/br8n && cp -r br8n/template/brain ~/my-brain

Claude Desktop / Claude Code / Cursor (any MCP client) — add:

json

{ "mcpServers": { "br8n": { "command": "npx", "args": ["-y", "@br8n/mcp", "/path/to/my-brain"] } } }

Then ask the model something the brain knows. It answers from the file and names it. Switch the model; same answer, same file.

Why files

Portable. cp -r brain/ new-machine/ is the whole migration. If you can't do that, you don't own it.
Inspectable. You can read every byte the model reads.
Model-agnostic. The folder doesn't care which model is on the other side of MCP.
It can push back. A decision stored with its why lets a model say "this conflicts with what you decided in March." A chat history can't.

The method is public on purpose

The layout and this server are MIT. What br8n charges for is hands on a real operation: eliciting what's actually in people's heads, shaping retrieval for the role, running it, and keeping it current. If you'd rather do it yourself, start here — most people should. The free first course is at br8n.io/lab.

Not affiliated

br8n on PyPI (a context-capture engine by a different author) is not this project.

Publishing (maintainers)

Configure npm's trusted publisher once for @br8n/mcp:

Publisher: GitHub Actions
Organization: Branded-Mayhem-Collective-LLC
Repository: br8n
Workflow: publish.yml
Allowed action: npm publish

Publish the version already committed to main, then update the official MCP Registry:

bash

gh workflow run publish.yml --ref main -f version=0.1.2
gh run watch --exit-status

mcp-publisher login github
mcp-publisher publish

The GitHub-hosted workflow uses npm trusted publishing (OIDC), so it requires no long-lived npm token. Change the version argument for future releases.

io.br8n/mcp