MCP-ready handoff memory

Keep your AI work moving without retelling the whole story.

Cerebro stores concise task memory by project so Claude, Codex, Gemini, and other agents can pick up where the last one stopped.

Memory records
Project-scoped
Agent access
Scoped tokens
Search
Fast retrieval
Live handoff snapshot
Recent memory
Claude to Codex
Refactor MCP token flow

Stored summary includes auth decisions, schema updates, and next steps for implementation.

Agent registry
Claudeactive
Codextoken scoped
Why it matters

The memory layer sits between sessions so the user explains intent once, not every time they switch tools.

Core product pillars

A memory system designed for agent transitions, not just note-taking.

01

Project memory ledger

Keep summaries anchored to repositories, tasks, and decision history instead of scattering them across chats.

02

Scoped MCP access

Issue tokens per agent client so memory retrieval stays precise and controlled.

03

Portable handoff formats

Import and export summaries as JSON or Markdown for backup, migration, and human review.

Step 1
Capture

An agent or user stores a concise task summary with project context.

Step 2
Scope

The memory is tied to a workspace and protected by token permissions.

Step 3
Retrieve

A new agent queries the same project and receives the relevant summaries.

Step 4
Continue

The next session resumes with decisions, status, and next actions already in place.

Self-hosted runtime

Run Cerebro on your own infrastructure.

The MVP ships with Docker Compose for Next.js and the Convex services, while public traffic is expected to flow through a host-level Caddy proxy. SQLite keeps private self-hosting simple today, while the deployment docs include the later move to Postgres if the memory volume grows.

Included surfaces
  • Public landing and docs page
  • Private dashboard and project workspace
  • MCP endpoint with scoped access tokens
  • JSON and Markdown portability