# Loci MCP server

> Provides AI agents with persistent memory using hybrid vector search, full-text retrieval, and a knowledge graph backed by PostgreSQL.

- Repository: https://github.com/tobs-code/loci-mcp
- Page: https://hostedmcp.org/servers/tobs-code/loci-mcp
- Transport: —
- Stars: 1
- License: Apache-2.0

## Tools (24)

- `remember`
- `remember_batch`
- `update_memory`
- `update_batch`
- `merge_memories`
- `forget`
- `recall`
- `recall_batch`
- `compile_context`
- `list_memories`
- `get_entity`
- `get_entities`
- `list_entities`
- `list_projects`
- `init_context`
- `unrelate`
- `find_connections`
- `find_paths`
- `list_relations`
- `contradictions_report`
- `verify_integrity`
- `prune_memories`
- `drop_project`
- `server_health`

## FAQ

### How does Loci handle conflicting information?

It features self-cleaning capabilities where contradictions are detected and the outdated fact is automatically superseded.

### Does Loci support GPU acceleration for embeddings?

Yes, embeddings run locally and will use an NVIDIA GPU if the Docker container has access to one, otherwise falling back to CPU.

### Can I use local LLMs to keep my memory content private?

Yes, by using providers like ollama or openai-compatible with a local endpoint, you can keep memory content on your machine.

### What kind of search capabilities does the server offer?

It provides hybrid retrieval, combining vector similarity with full-text search to find both paraphrases and exact identifiers.
