Collective Memory
Store and semantically recall project facts across Claude/Cursor sessions so the agent does not re-discover the same context every day.
Overview
collective-memory is an MCP server for the Build phase that provides persistent semantic memory across AI coding sessions using embeddings.
What is this MCP server?
- Persistent semantic memory across AI coding sessions
- Embedding-backed recall via required OPENAI_API_KEY
- npm collective-memory v0.1.2 stdio MCP transport
- Reduces repeated re-explanation of stack choices and conventions
- Early 0.1.x release—expect evolving APIs and storage behavior
- Server version 0.1.2
- Required secret env: OPENAI_API_KEY
- Transport: stdio npm package collective-memory
What problem does it solve?
Every new agent session forgets your stack, decisions, and prior fixes, so you waste time re-teaching the same project context.
Who is it for?
Solo builders running long-lived agent sessions who want OpenAI-embedding-backed recall without building a custom memory layer.
Skip if: Teams that cannot use OpenAI embeddings, need enterprise audit-grade memory stores on day one, or want zero external API dependencies.
What do I get? / Deliverables
The agent can store and retrieve semantic memories across sessions so ongoing build work stays aligned with how you already decided to ship.
- Semantic memory store queried from the agent across sessions
- Less repeated project onboarding in every new chat
- Durable recall of decisions, conventions, and solved issues
Recommended MCP Servers
Journey fit
Long build arcs lose thread when every session starts cold; persistent memory belongs alongside the agents and integrations you ship. Agent-tooling is the shelf because the server augments the coding agent runtime with cross-session recall, not customer-facing product features.
How it compares
Cross-session agent memory MCP, not an in-repo docs skill or vector DB admin console alone.
Common Questions / FAQ
Who is collective-memory for?
Solo developers using MCP agents who need project context to persist across sessions without manually curating a giant CLAUDE.md every time.
When should I use collective-memory?
Use it once a codebase or product direction stabilizes enough that repeating conventions, pitfalls, and decisions to the agent becomes painful.
How do I add collective-memory to my agent?
Set OPENAI_API_KEY in the MCP environment, install npm package collective-memory, add stdio server config, restart the agent, and use memory tools to save and query notes.