MCP Memory Service
Self-hosted long-term memory for agents: semantic search, a knowledge graph and autonomous consolidation over MCP and a REST API
- 2,117 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Runs a local service that reads and stores project data
- In remote mode it opens network access and needs authentication setup
Install
- Works in
Manual install
pip install mcp-memory-serviceInstalls the package from PyPI. Then configure your MCP client.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Runs a local service that reads and stores project data
- In remote mode it opens network access and needs authentication setup
This is third-party code. Review the repository files before installing.
What it does
The service stores an agent's memory across sessions so you stop re-explaining architecture and past decisions. It keeps notes in a local vector-search database, links them into a knowledge graph with typed edges and can compress older entries on its own. Access runs over MCP for agents such as Claude Code, Claude Desktop and Cursor, and over a REST API with many endpoints for pipelines built on LangGraph, CrewAI, AutoGen and any HTTP client. Embeddings are computed locally via ONNX, so data stays on your infrastructure and no external keys are needed for that. Several storage backends are available: SQLite for a single user, Cloudflare and a hybrid mode for multi-device sync, and Milvus as a dedicated vector database. A remote mode with OAuth and HTTPS enables use from the claude.ai browser.
Who it is for. For developers and devops engineers who need shared long-term memory for agents and pipelines on their own infrastructure.
Good fit when
- You want an agent to remember project context and decisions across sessions
- Several agents or pipelines need to share one memory store
- You want memory on your own infrastructure, without cloud lock-in or per-request cost
Not a fit when
- You want ready cloud memory without deploying and maintaining a service
- The agent's built-in memory is enough and a separate store is unnecessary
Example request
Save to memory that our API rate limit is 100 requests per minute and recall it next sessionMCP
- Transport
- stdio, http, sse
- Authentication
- not required
Environment variables
MCP_ALLOW_ANONYMOUS_ACCESS- Allows anonymous access to the HTTP mode for local pipelines
MCP_OAUTH_ENABLED- Enables OAuth for remote access from the claude.ai browser
Limitations
This is a separate service you install and maintain: SQLite is enough locally, while a team and sync need Cloudflare, Milvus or Docker. Browser remote access needs HTTPS and OAuth setup, and exposing it publicly needs a tunnel. Development has moved to GitHub, and the copy on the old host is no longer updated. Apache-2.0 licensed.
How to disable. Remove the memory block from the MCP client config, or for Claude Code run claude mcp remove memory. Stop the service process and, if you wish, remove the package: pip uninstall mcp-memory-service.
FAQ
Does data go to the cloud?
Not by default. Embeddings run locally via ONNX and the database stays with you. Cloudflare and Milvus cloud backends are optional for sync.
Which agents does it work with?
Any MCP client: Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Gemini CLI, Codex CLI, OpenCode, Goose and others. Pipelines use the REST API.
Discussion
No questions or reviews yet.
Related
MCP serversAgent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
A self-hosted MCP memory server that keeps preferences and decisions across conversations in Claude Code, Cursor and other clients
A self-hosted MCP server for long-term agent memory: structured records, versioned rollback and a visual dashboard
A long-term memory MCP server for agents: one Rust binary that runs offline and never calls a language model
Membase keeps one shared memory for AI agents that Claude Code, Cursor and Codex read through MCP, an SDK or the membase command
An MCP server and API of shared memory for several agents: facts, decisions and events live in one database with semantic search.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| MCP Memory Service (this card) | no | not required | 2,117 | |
| memory-lancedb-pro | no | 4,450 | a long-term memory plugin for the OpenClaw agent on LanceDB | |
| Lossless Claw | no | 4,924 | an OpenClaw plugin that keeps the whole conversation in SQLite | |
| IWE | no | not required | 1,823 | iWE keeps project memory in linked Markdown notes |
README in short
The README presents an open-source memory backend for agents accessible via MCP, a REST API, OAuth, a CLI and a dashboard. Installation is pip install mcp-memory-service, then the client is configured: a config block for Claude Desktop, the claude mcp add command for Claude Code. For LangGraph, CrewAI, AutoGen and HTTP clients you run an HTTP mode with the REST API. Embeddings run locally via ONNX, and SQLite, Cloudflare, hybrid and Milvus backends are supported. A remote mode serves claude.ai through a tunnel and OAuth. Apache-2.0 licensed, published on PyPI.