# RepoMemory: a shared ledger so Cursor stops redoing changes your team already reverted (remote MCP)

**URL:** <https://forum.cursor.com/t/repomemory-a-shared-ledger-so-cursor-stops-redoing-changes-your-team-already-reverted-remote-mcp/176792>\
**Category:** Built for Cursor\
**Tags:** mcp, rules\
**Created:** [October 30, 2026, 1:25pm UTC](https://forum.cursor.com/t/repomemory-a-shared-ledger-so-cursor-stops-redoing-changes-your-team-already-reverted-remote-mcp/176792 "2026-10-30T13:25:58Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![dimaino](https://avatars.discourse-cdn.com/v4/letter/d/3ec8ea/32.png) [@dimaino](https://forum.cursor.com/u/dimaino)\
**Post date:** [October 30, 2026, 1:25pm UTC](https://forum.cursor.com/t/repomemory-a-shared-ledger-so-cursor-stops-redoing-changes-your-team-already-reverted-remote-mcp/176792/1 "2026-10-30T13:25:58Z")

</div>

Disclosure: I made this. It’s a hosted MCP server with a free tier (300 entries a month, no card), and paid plans above that.

**The problem:** project rules are great for “how we write code here”, but they don’t hold history. Say someone adds a captcha to checkout, it cuts completed orders, and it gets reverted. Weeks later the agent is asked to “stop bots on checkout” and adds it straight back, because the reason for the revert lives in a PR comment it never sees. Teammates on Claude Code or Codex have the same gap, each with their own rules file.

**What RepoMemory does:** it keeps one ledger per repo of decisions, lessons, guardrails and tasks, which every agent on the team reads and writes over MCP:

- `brief`: every guardrail plus the open tasks, at the start of a session
- `recall`: keyword search before planning a change (“captcha checkout” finds that lesson)
- `record_entry`: the agent writes down what it learned, or why something was reverted
- `claim_task` / `retire_entry`: a small task queue, so two agents don’t take the same thing

Search is plain keyword ranking, not embeddings, so you can predict what comes back and read the whole ledger yourself.

**Setup in Cursor:** run `npx @plexsicity/repomemory init` in the repo (it writes `.cursor/mcp.json`), or add it by hand with a key from the dashboard:

```json
{
  "mcpServers": {
    "repomemory": {
      "url": "https://api.repomemory.dev/mcp",
      "headers": { "X-API-Key": "<your key>" }
    }
  }
}

```

Then add a short rule telling the agent to call `brief` at the start and `recall` before changes. Docs, with the exact rule text: [API Reference · RepoMemory](https://repomemory.dev/docs)

I also wrote up how this fits next to project rules and Team Rules: [Cursor rules for a team: project rules, team rules and shared memory · RepoMemory](https://repomemory.dev/guides/cursor-rules-for-teams)

I’d like feedback from teams in particular: does keyword recall hold up for you, and what would you need before pointing it at a shared repo?