Memwright
Embed agent memory in your stack with SQLite, pgvector, and Neo4j graph search exposed over MCP.
Overview
Memwright is an MCP server for the Build phase that embeds agent memory across SQLite, pgvector, and Neo4j with semantic and graph search.
What is this MCP server?
- SQLite for lightweight local persistence alongside pgvector semantic search
- Neo4j graph search for relationship-heavy agent recall
- Semantic search via OpenRouter or OpenAI embedding API keys
- Stdio MCP PyPI package memwright 0.1.3 with documented default connection strings
- Configurable PG_CONNECTION_STRING and NEO4J_PASSWORD for local or dockerized stacks
- PyPI memwright version 0.1.3 with stdio MCP transport
- Default Postgres URL postgresql://memwright:memwright@localhost:5432/memwright documented in server metadata
Community signal: 13 GitHub stars.
What problem does it solve?
Agents forget structured relationships and vector nuance when memory is only a flat file or chat transcript.
Who is it for?
Builders already running Postgres or Neo4j locally who want graph-plus-vector recall in their coding agent.
Skip if: Anyone seeking zero-setup cloud memory with no database containers or embedding API keys.
What do I get? / Deliverables
After registration, your agent queries hybrid SQL, vector, and graph stores through one MCP tool surface.
- MCP-accessible hybrid memory spanning relational, vector, and graph layers
- Repeatable local dev topology for agent recall testing
Recommended MCP Servers
Journey fit
How it compares
Multi-store embedded memory MCP, not Attestor-style bi-temporal audit replay.
Common Questions / FAQ
Who is Memwright for?
Solo builders and small teams wiring coding agents to real databases who need semantic search and graph traversal without a managed memory vendor.
When should I use Memwright?
Use it when agent recall must span vectors and entity relationships, or when you want SQLite for dev and pgvector for production-like tests.
How do I add Memwright to my agent?
Install memwright from PyPI, set PG_CONNECTION_STRING and NEO4J_PASSWORD, add OPENROUTER_API_KEY or OPENAI_API_KEY for embeddings, register stdio MCP in your agent client.