Memora
An MCP layer of persistent collective memory for agents: structured storage, semantic search, a relation graph and a supersession lineage
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Stores accumulated work memory in a local database
- Absorb and deduplication call an LLM
Install
Manual install
pip install memora-mcpFor semantic search: pip install "memora-mcp[local]". Then connect the server to the agent as MCP.
This is third-party code. Review the repository files before installing.
What it does
Memora gives agents a shared persistent memory through MCP. Facts can be fed in with absorb: an LLM classifies each against the store as duplicate, update, contradiction, related or new, skips duplicates and links relations, with a dry_run preview. Updates do not delete old knowledge but supersede it, and retrieval follows the chain to the current version by default. A memory_digest by topic bundles relevant memories, open TODOs, related edges and source IDs. The store is SQLite with optional cloud sync, plus semantic search, a knowledge graph and structured document storage.
Who it is for. For developers and teams who need shared agent memory with a relation graph and update history.
Good fit when
- You need shared memory across several agents and sessions
- You want updates to supersede old knowledge rather than overwrite it
- You need semantic search and a relation graph over memory
Not a fit when
- A simple text notes file is enough
- You do not need fact classification and a graph, only a quick cache
Example request
Absorb the results of this sprint into memory and later build a digest on the database migration topicLimitations
The base install is via pip, and local embeddings and some features need extra dependencies or keys. Absorb and deduplication rely on an LLM, so they need model access. The canonical store is SQLite, and cloud sync is optional and configured separately.
How to disable. Remove the MCP entry from the agent config and uninstall the memora-mcp package; the SQLite database file can be deleted manually.
MCP
- Transport
- stdio, http
- Authentication
- not required
Security check
- Stores accumulated work memory in a local database
- Absorb and deduplication call an LLM
README in short
The README describes Memora as an MCP layer of persistent collective memory for agents. The store is SQLite with optional cloud sync and multi-database routing through one process. Absorb classifies and links facts, updates supersede old knowledge via a supersession lineage, and memory_digest bundles memories, TODOs, edges and sources by topic. There is embedding-based semantic search, full-text and hybrid search, a knowledge graph with visualization and structured document storage as fragment trees. Install is via pip, with options for local embeddings.
FAQ
Are old records deleted on update?
No. An update supersedes old knowledge, and retrieval follows the chain to the current version by default, with full history available.
What does absorb do?
It classifies each fact against the store as duplicate, update, contradiction, related or new, with a dry_run preview.
Related
A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything