<div align="center">

Context Fabric

Persistent memory for AI coding agents. Your agent remembers everything -- across sessions, projects, and tools.

</div>

[!NOTE] Beta Software. Context Fabric works and is actively used, but APIs and storage formats may change between versions. Pin your version and check the CHANGELOG before upgrading.

---

The Problem

Every time an AI CLI session ends, its context vanishes. Decisions, patterns, bug fixes -- gone. Next session, you start from scratch.

The Solution

Context Fabric is an MCP server that gives your AI agent a three-layer memory system and time-aware orientation. It remembers what happened, what changed while you were away, and what matters right now. No external APIs. No cloud. Everything runs locally.

Features

• Three-layer memory -- Working (L1), Project (L2), Semantic (L3). Memories auto-route to the right layer.
• Local code indexing -- Scans source files, extracts symbols (functions/classes/types), and stays up-to-date via file watching. Search by text, symbol name, or semantic similarity.
• Semantic recall -- Search by meaning using in-process vector embeddings. No API keys needed.
• Time-aware orientation -- "What happened while I was away?" Offline gap detection, timezone support, session continuity.
• Ghost messages -- Relevant memories surface silently without cluttering the conversation.
• Pattern detection -- Auto-captures and reuses code patterns across projects.
• Self-installing -- Ask your AI to run context.setup and it configures itself into any supported CLI.
• Docker-first -- Cross-platform docker run --rm -i. No Node.js required on the host.
• 16 MCP tools -- Store, recall, orient, time, summarize, promote, ghost, patterns, events, searchCode, setup.
• Zero external dependencies -- All storage is SQLite. All search is local. Nothing leaves your machine.

Supported CLIs

CLI	Setup	Docs
Claude Code	context.setup({ cli: "claude-code" })	Guide
Kimi	context.setup({ cli: "kimi" })	Guide
OpenCode	context.setup({ cli: "opencode" })	Guide
Codex CLI	context.setup({ cli: "codex" })	Guide
Gemini CLI	context.setup({ cli: "gemini" })	Guide
Cursor	context.setup({ cli: "cursor" })	Guide
Claude Desktop	context.setup({ cli: "claude" })	Guide

[!TIP] Skip manual config entirely. Once Context Fabric is running in any CLI, the AI can install itself into all the others -- see Quick Start step 3.

Quick Start

Get running in 3 steps:

# 1. Clone and build the Docker image (~2 min)
git clone https://github.com/Abaddollyon/context-fabric.git
cd context-fabric
docker build -t context-fabric .

# 2. Test that it works
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' \
  | docker run --rm -i context-fabric

3. Add to your CLI. Point your MCP config at the Docker transport:

docker run --rm -i -v context-fabric-data:/data/.context-fabric context-fabric

See CLI Setup for copy-paste configs for all 7 CLIs, or let the AI do it -- once Context Fabric is running in one CLI, tell it:

"Install and configure Context Fabric for Cursor using Docker"

It writes the config automatically. No manual editing needed.

<details> <summary><strong>Local install (without Docker)</strong></summary>

Requires Node.js 22.5+:

git clone https://github.com/Abaddollyon/context-fabric.git
cd context-fabric
npm install && npm run build

The server is at dist/server.js. Point your CLI's MCP config at node dist/server.js.

</details>

What It Looks Like

Start a session. The AI calls context.orient and instantly knows where it is:

It is 9:15 AM on Wednesday, Feb 25 (America/New_York).
Project: /home/user/myapp.
Last session: 14 hours ago. 3 new memories added while you were away.

Store a decision. The AI remembers it next session, next week, across tools:

// Store
{ "type": "decision", "content": "Use Zod for all API validation. Schemas in src/schemas/." }

// Recall (semantic search -- doesn't need exact words)
{ "query": "how do we validate inputs?" }
// => "Use Zod for all API validation. Schemas in src/schemas/." (similarity: 0.91)

No configuration. No prompting. Memories route to the right layer automatically.

How It Works

CLI (Claude Code, Cursor, etc.)
  |
  | MCP protocol (stdio / Docker)
  v
Context Fabric Server
  |-- Smart Router -----> L1: Working Memory  (in-memory, session-scoped)
  |-- Time Service        L2: Project Memory  (SQLite, per-project)
  |                       L3: Semantic Memory  (SQLite + vector search, cross-project)

Memories auto-route to the right layer. Scratchpad notes go to L1 (ephemeral). Decisions and bug fixes go to L2 (persistent). Code patterns and conventions go to L3 (searchable by meaning). See Architecture for the full deep dive.

Documentation

Resource	Description
Getting Started	Installation, first run, Docker and local setup
CLI Setup	Per-CLI configuration (all 7 supported CLIs)
Tools Reference	All 16 MCP tools with full parameter docs
Memory Types	Type system, three layers, smart routing, decay
Configuration	Storage paths, TTL, embedding, environment variables
Agent Integration	System prompt instructions for automatic tool usage
Architecture	System internals, data flow, embedding strategy
Changelog	Version history and migration notes

Contributing

Contributions are welcome. See CONTRIBUTING.md for how to get started.

License

MIT

---

<div align="center">

Stop re-explaining your codebase every session.

Get Started | View All Tools | Report a Bug

</div>