# MemoryPilot MCP server

> Gives AI clients persistent local long-term memory with hybrid semantic search, auto-classification and compressed context recall.

- Repository: https://github.com/Soflutionltd/MemoryPilot
- Page: https://hostedmcp.org/servers/soflutionltd/memorypilot
- Transport: stdio
- Stars: 18
- Latest version: 4.13.0

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `MEMORYPILOT_CROSS_RERANK` | no | no |  |
| `MEMORYPILOT_CROSS_RERANK_WEIGHT` | no | no |  |

## Tools (8)

- `add_memory`
- `add_memories`
- `add_transcript`
- `ingest_session`
- `get_memory`
- `update_memory`
- `delete_memory`
- `list_memories`

## FAQ

### What does MemoryPilot do for AI assistants?

It gives AI clients persistent, searchable memory with project awareness and semantic understanding so context survives between sessions.

### Does MemoryPilot require an internet connection or external API calls?

No, it runs as a single binary with zero runtime dependencies and zero API calls.

### How much does MemoryPilot reduce token usage?

Built-in AAAK compression and memory capsules cut token consumption by 5-10x when loading context.

### How does MemoryPilot search memories?

It uses hybrid search combining BM25 with jina embeddings via reciprocal rank fusion plus cross-encoder reranking.

### Which AI clients can install MemoryPilot?

It offers one-liner installs for Cursor, Claude Desktop, VS Code, Windsurf, Claude Code, Codex and ChatGPT.