Team Brain MCP

Collaborative AI conversations for teams, using Git as the sync backbone.

Each team member uses their own AI subscription (Claude, ChatGPT, etc.). The MCP server runs locally, manages a shared Git repo where conversations are stored as structured markdown, and keeps everything in sync transparently.

How It Works

┌──────────────────────────────────────────────┐
│  Your AI Client (Claude, ChatGPT, etc.)      │
└──────────────────┬───────────────────────────┘
                   │ MCP protocol
┌──────────────────▼───────────────────────────┐
│  Team Brain MCP Server (runs locally)        │
│  - Git sync on every operation               │
│  - Conversation serialize/deserialize        │
└──────────────────┬───────────────────────────┘
                   │ git operations
┌──────────────────▼───────────────────────────┐
│  Shared Git Repository (GitHub/GitLab/etc.)  │
│  - One folder per conversation               │
│  - Append-only markdown files                │
└──────────────────────────────────────────────┘

One person brainstorms with AI, commits the conversation, and others can continue it — adding their own questions and insights. The result is a growing, attributed, multi-person conversation any AI client can load and continue.

Quick Start

1. Install

npm install -g team-brain-mcp
# or from source:
git clone https://github.com/zonder/team-brain-mcp
cd team-brain-mcp && npm install && npm run build

2. Set Up a Shared Repo

# Create or clone your team's conversation repo
git clone https://github.com/your-team/team-brain.git ~/team-brain

3. Configure

Set the repo path via environment variable:

export TEAM_BRAIN_REPO=~/team-brain

Or create ~/.team-brain/config.json:

{
  "repo": "/home/you/team-brain"
}

4. Add to Your AI Client

Add to your MCP client config (e.g., Claude Desktop claude_desktop_config.json):

{
  "mcpServers": {
    "team-brain": {
      "command": "node",
      "args": ["/path/to/team-brain-mcp/dist/index.js"],
      "env": {
        "TEAM_BRAIN_REPO": "/path/to/team-brain"
      }
    }
  }
}

MCP Tools

| Tool | Description | |------|-------------| | list_conversations | Show all team conversations, filterable by status | | open_conversation | Load a conversation so the AI can continue it | | share_conversation | Save the current AI chat as a new shared conversation | | commit_changes | Append new messages and push to the team | | add_context | Add a supporting file (PDF, doc, data) to a conversation | | set_status | Mark a conversation as active, pending_review, or archived | | check_updates | See what the team has been up to since you last synced |

Conversation Format

Conversations are stored as structured markdown in conversation.md files:

---
title: Feature opportunities Q2
status: active
created: 2026-03-28T14:30:00Z
started_by: Alex Rivera <[email protected]>
---

<!-- msg:001 | human | Alex Rivera <[email protected]> | 2026-03-28T14:30:00Z -->

What if we made onboarding completely self-serve?

---

<!-- msg:002 | assistant | 2026-03-28T14:30:45Z -->

Self-serve onboarding needs three things: a setup wizard, sample data, and a clear path to value...

---

<!-- msg:003 | human | Jamie Nguyen <[email protected]> | 2026-03-29T09:15:00Z -->

Building on Alex's idea — what about product-qualified signals for upsell?

---

• Human-readable — open in any editor or on GitHub
• Machine-parseable — the MCP server converts it to LLM message arrays
• Attributed — every message shows who wrote it and when
• Append-only — new messages go at the end, making Git merges reliable

User Identity

Derived from your Git config — no separate setup needed:

git config user.name   # "Alex Rivera"
git config user.email  # "[email protected]"

Example Repo Structure

team-brain/
├── feature-opportunities/
│   ├── conversation.md
│   └── context/
│       └── product-roadmap.md
├── api-redesign/
│   ├── conversation.md
│   └── context/
│       └── current-api-spec.yaml
└── .gitignore

See the example-repo/ folder for working examples.

Design Principles

1. No shared database — Git is the only shared state
2. No API costs — AI compute runs on each user's own subscription
3. Always in sync — every tool call starts with git pull --rebase
4. Append-only — messages are never edited, making merges reliable
5. LLM-agnostic — the markdown format works with any AI provider

License

MIT