OzBrain for Cursor
Your rules files stop being copies.
Updated
OzBrain connects to Cursor as an MCP server with the MCP URL. After you approve once, the agent in every repo reads your routing index, pulls the articles a task needs, and stages what it learns so Claude, ChatGPT, and Claude Code share the same facts. Rules files shrink to what is truly repo-specific.
The knowledge you keep copying between projects stops being copied. Grant the brain to a teammate and their agents read it too.
One MCP entry, one approve, every repo.
- 01
Add the server
In Cursor: Settings, then MCP, then add a new server with the MCP URL.https://ozbrain.com/api/mcp - 02
Sign in and approve
Cursor opens a browser window. Sign in to OzBrain and approve the connection once. Cursor then refreshes the session on its own while the connection stays healthy. - 03
Work
The brain’s tools appear to the agent. At the start of a task it reads the routing index and pulls the articles the work needs.
What the Cursor agent can do once connected
The connector exposes your brain as tools. The agent decides when to use them, and every write is attributed in the article history.
Read the routing index
A compact listing of every article and what it covers, so the agent finds knowledge by structure instead of whatever made it into the rules file.
Start any repo with full context
Fixing a client integration, it reads the client article and your stack preferences. New project, same brain: day-one context with nothing copied over.
Stage what it learns
Decisions and gotchas worth keeping get staged and routed mid-build, so they survive the session and reach every other agent you use.
Surface conflicts before they land
If a write contradicts what the brain holds, the write pauses and the agent sees the conflict. It can record an exception; a different agent or you must approve a supersede.
Configured once, works in every repo
The connector lives at the Cursor level, not the project level, so a fresh clone has the same brain behind it as your oldest repo. Rules files shrink to build commands and local quirks.
Keep Cursor. Connect the brain under it.
Rules files stay for repo quirks. OzBrain holds the knowledge that repeats across projects, agents, and people you grant access to.
Create your brain, add the MCP URL in Cursor MCP settings, approve once, then ask the agent to read the routing index.
Questions people ask before connecting
Does this replace .cursor/rules?
No, it shrinks them. Repo-specific instructions stay in the rules file; knowledge that repeats across projects moves to the brain. The full argument is at OzBrain vs CLAUDE.md files.
Is this the same brain as my chat agents?
Yes. What the Cursor agent stages, Claude and ChatGPT can read, and the other way around. One brain, every surface, each write attributed to the agent that made it.
Does it slow the agent down?
Reads happen when the agent decides it needs knowledge, not on every request. The routing index is compact by design, so knowing what exists costs little and reading deeply is reserved for the articles a task touches.
Why did every Cursor agent ask me to reauthenticate at once?
OzBrain ends a shared auth session when a stale refresh token is reused after rotation. In Cursor that can show up as every open agent demanding reauth together, not one session at a time. A longer access token reduces how often refresh runs; it does not guarantee uninterrupted multi-agent use if a stale refresh token is presented. If the MCP URL used to be on ozbrain.co, remove the old server entry and add https://ozbrain.com/api/mcp again.
Do I ever need to sign in again?
Yes, when the connection is revoked, the client loses its refresh token, or OzBrain ends the session after a bad refresh. That is reauthentication, not OzBrain being down. Keep the server URL at https://ozbrain.com/api/mcp