Context Post: shared memory across agents

Give Cursor, Claude Code and Codex one shared project context so new sessions stop forgetting

Problem

When building with Cursor, Claude Code and Codex, the background and decisions agreed in one window vanish when you switch tools or start a new session — so you re-explain everything or maintain an ever-growing MEMORY.md. Each agent keeps its own memory, forcing users to manually ferry constraints and decisions between tools.

Solution

An agent-neutral project context space that stores source material, decisions, constraints and methods in one place and exposes them over MCP to multiple agents. Agents can write new conclusions back after authorization, conflicts and sensitive content require confirmation, and everything is exportable.

Users

Developers juggling multiple AI coding tools, tech leads at small teams, and AI-assisted writers and researchers

MVP

One context library per project with manual edits and versioning; an MCP server that Cursor/Claude Code can read; confirmation before write-back; conflict prompts; one-click Markdown export

Revenue

Subscription by project count or seat (free for one project, paid for multi-project and team sharing)

Why now

Parallel agents are now routine and MCP makes cross-tool read/write feasible, yet official memory stays siloed; a neutral third party is something no single model vendor can credibly be