Shared team memory for engineering organizations

Every engineer runs the open-source brain locally — it distills coding sessions into durable memories, and the raw transcripts never leave their machine. Only distilled, team-safe memories sync to your workspace. From there, anyone on the team can search that knowledge from any MCP-capable coding agent — Claude, Codex, Cursor and more — across languages.

How it works

Distill locally

The OSS brain turns each session into concise memories on your machine — decisions, fixes, gotchas.

Sync automatically

Team-safe memories push to your workspace as they are saved. Private ones stay local, always.

Ask from any agent

Point any MCP-capable agent at the remote server and search the whole team's memory, cross-language.

Connect

Plug the team memory into the coding agents you already use. Two paths — most people want both:

Contribute

Install the plugin — your memories flow to the team automatically.

Consume

Point any agent at the remote MCP — zero install.

Install the plugin

npm install -g brain-rag

Works with Claude & Codex out of the box as your local second brain.

Your API key

Every member has a personal key (brk_…) shown once at signup or invite. If you don't have one, ask your workspace owner for an invite, or create a workspace.

The key identifies you and your workspace — one key = one person + one workspace.

Every snippet below uses the placeholder brk_YOUR_KEY — replace it with your own key. Real keys are never displayed in this dashboard.

Link it to your workspace

brain-rag cloud login https://rag.optimusdatalab.com    # paste your brk_ key

That's it. From now on sync is automatic — new team-safe memories push as they are saved.

private:true memories never leave your machine.
Shared something by mistake? Retract it from the dashboard — it's removed for everyone.
Prefer manual control? brain-rag cloud auto off switches to manual sync, and brain-rag cloud sync --review dry-runs what would leave this machine.

Privacy

Only distilled Layer-2 memories sync — raw transcripts never leave the author's machine, and memories flagged private:true stay local too.

Point your agents at the remote MCP

Querying needs no install — any MCP-capable agent can read the team memory directly.

Claude

claude mcp add --transport http brain-team https://rag.optimusdatalab.com/mcp \
  --header "Authorization: Bearer brk_YOUR_KEY"

Add --scope user to make it available in every repo. Then in any session ask e.g. “has anyone on the team dealt with payment retries?” — Claude calls search_team_context.

Claude Desktop / claude.ai

Custom connectors want OAuth, so use the mcp-remote bridge for header auth:

claude_desktop_config.json

{
  "mcpServers": {
    "brain-team": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://rag.optimusdatalab.com/mcp", "--header", "Authorization: Bearer brk_YOUR_KEY"]
    }
  }
}

Codex CLI

~/.codex/config.toml

[mcp_servers.brain-team]
command = "npx"
args = ["-y", "mcp-remote", "https://rag.optimusdatalab.com/mcp", "--header", "Authorization: Bearer brk_YOUR_KEY"]

Newer Codex builds also support HTTP MCP servers natively (url + headers); the bridge form above works everywhere.

Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "brain-team": {
      "url": "https://rag.optimusdatalab.com/mcp",
      "headers": { "Authorization": "Bearer brk_YOUR_KEY" }
    }
  }
}

Two tools, cross-language search

search_team_context
Hybrid semantic + exact search over the team store — works cross-language.
explore_team_memory
Fetch one memory with provenance and supersede chain.

Try it: ask your agent a question only a teammate's memory can answer.