MCP Memory Server  ·  Open Source

Your repos,
remembered.

Fossel is a local MCP server that gives your AI tools persistent memory about every codebase you contribute to. Reviewer patterns, bug fixes, conventions — never explain the same thing twice.

Get started free Star on GitHub
fossel · cursor chat
you → work on issue #892 in RocketChat
fossel → loading context for RocketChat...
⟡  Uses Blaze, not React. Components in client/
⟡  Reviewer Guilherme prefers PRs under 200 lines
⟡  Issue #892 — tried RoomHeader.js, try Toolbar.js next
✓  3 memories loaded. Picking up where you left off.

AI is smart.
But painfully forgetful.

Every time you open a new chat to work on a PR, your AI assistant starts from zero. You re-explain the project conventions. You re-describe the reviewer preferences. You re-trace the same bug you already solved last week.

Open-source contributors switch between 5, 10, even 20 repos. Each one has its own culture, its own quirks, its own people. No AI tool remembers any of it across sessions.

Fossel fixes this. Permanently.

Simple by design.
Powerful in practice.

01

Install in 30 seconds

Run npx fossel and add one JSON snippet to your Cursor or Claude config. No accounts, no cloud, no data leaving your machine.

02

Memories build automatically

Tell your AI what you learned — reviewer preferences, bug fixes, architectural decisions. Fossel stores them locally in SQLite with semantic search.

03

Context loads on every chat

Next time you open a chat about a repo, Fossel automatically surfaces the most relevant memories. Pick up exactly where you left off, every time.

Every kind of knowledge,
preserved.

 convention

Coding Conventions

Uses Blaze, not React. Components live in client/. Avoid TypeScript generics in PR descriptions.

 RocketChat/Rocket.Chat
 reviewer_pattern

Reviewer Preferences

Guilherme Gazzo prefers PRs under 200 lines. Sing Li always asks for edge case tests. Don't merge without approval from both.

 RocketChat/Rocket.Chat
 bug_fix

Bug Fixes & Root Causes

Subscription calls in RoomHeaderToolbar need debouncing — fires 3× on re-render without it. Fixed with 200ms delay.

 RocketChat/Rocket.Chat
 decision

Architectural Decisions

Decided NOT to use Prisma here — too much overhead for the read-heavy query pattern. Sticking with raw SQL.

 expressjs/express
 issue

Issue Context

Issue #892 — tried fixing in RoomHeader.js but conflicted with i18n module. Next step: try RoomHeaderToolbar.js instead.

 RocketChat/Rocket.Chat
 export

Team Memory Sharing

Export your repo knowledge as JSON and share it with teammates. They import 6 months of your learnings in one command.

 Any repo · Any team

Up in 30 seconds.
Works everywhere.

1

Install via npx

No global install needed. Works with Node.js 18+. All data stored locally at ~/.fossel/memory.db.

2

Add to your AI tool config

One JSON snippet in Cursor, Claude Desktop, or any MCP-compatible tool. Restart once.

3

Start building memory

Tell your AI to save context. It calls Fossel's tools automatically — nothing manual required.

Terminal
$ npx fossel
✓ Fossel v1.0.0 started
  Memory DB: ~/.fossel/memory.db
  Transport: stdio (MCP ready)
~/.cursor/mcp.json
{
  "mcpServers": {
    "fossel": {
      "command": "npx",
      "args": ["-y", "fossel"]
    }
  }
}

Your tools.
All of them.

Cursor✓ ready
Claude Desktop✓ ready
VS Code (Copilot)✓ ready
Windsurf✓ ready
Continue.dev✓ ready
Zedcoming soon

Any tool that supports the Model Context Protocol works with Fossel.

Stop re-explaining.
Start building.

Free, open-source, and fully local. Your memories never leave your machine.