# Anchor MCP

> Persistent memory server for AI with drift detection to prevent hallucinations from stale data.

Anchor MCP is a Model Context Protocol (MCP) server that provides LLMs with long-term memory. It runs locally on your machine with zero cloud dependencies. Unlike standard vector stores, Anchor actively manages knowledge freshness using statistical drift detection to prevent retrieval of obsolete information.

## Installation

```
pip install git+https://github.com/Arkya-AI/anchor-mcp.git
anchor-mcp init
```

The `init` command auto-detects Claude Desktop, Claude Code, Cursor, and Windsurf, and registers Anchor with each one.

## Key Features

- Cross-session memory across all MCP clients
- 100% local — no API keys, no cloud, ~500MB disk, ~200MB RAM
- Temporal intelligence with exponential decay and access frequency boosting
- Drift detection that auto-flags stale memories when knowledge regions shift
- Source linking for traceable decisions
- Zero configuration — two commands to install

## How It Works

1. Local embeddings via all-MiniLM-L6-v2 (384-dim, CPU)
2. Voronoi partitioning with 16 frozen centroids for topic clustering
3. FAISS similarity search with temporal re-ranking
4. Drift detection pipeline: per-cell statistics, k-NN adjacency, Laplacian smoothing, adaptive thresholds

## 11 MCP Tools

### Store & Recall
- `anchor_store` — Save a named memory with importance level and tags
- `anchor_recall` — Semantic search with temporal scoring
- `anchor_deep_recall` — Recall + read source files for full context
- `anchor_learn` — Auto-capture facts, preferences, decisions from conversation

### Management
- `anchor_list` — List all stored memories, filter by tag
- `anchor_delete` — Remove a memory by ID
- `anchor_contradict` — Mark outdated memory stale, store corrected version

### Intelligence
- `anchor_auto` — Auto-retrieve relevant context at conversation start
- `anchor_inspect` — View Voronoi cell distribution and density
- `anchor_save_session` — Save session summary with source linking
- `anchor_drift_check` — Run drift detection to flag stale memories

## Storage

All data stored locally at `~/.anchor/` (JSON + SQLite + FAISS index).

## Requirements

- Python 3.10+
- macOS, Linux, Windows (WSL)

## Links

- GitHub: https://github.com/Arkya-AI/anchor-mcp
- Website: https://anchormcp.arkya.ai
- License: MIT
- Author: Arkya AI (https://github.com/Arkya-AI)