Noureddine RAMDI / Memcord: a privacy-first self-hosted MCP server for AI memory management

Created Mon, 04 May 2026 10:23:01 +0000 Modified Sat, 23 May 2026 20:41:27 +0000
958 Words
github-stars python mcp ai-memory self-hosted privacy summarization

ukkit/memcord

Memcord is a privacy-first, self-hosted Model Context Protocol (MCP) server designed to offer AI memory management capabilities without relying on cloud services. It tackles the common challenge of maintaining persistent conversational context across AI sessions by storing memory locally and isolating it per project or slot. This approach ensures that your AI workflows keep context consistent and private, without sending sensitive information to external APIs or cloud storage.

What memcord does and how it manages AI memory

At its core, memcord implements the MCP server protocol with a focus on local-first AI memory. MCP is a protocol that standardizes how AI models access external context resources, enabling modular and composable AI systems. Memcord acts as a backend server providing persistent memory context to AI models or agents.

The architecture uses a slot-based system to isolate memory contexts. Each slot represents a separate context or project workspace, enabling multiple AI workflows to coexist without interfering with each other’s memory. The repository uses Python 3.10+ with a lightweight Python package manager called uv to manage dependencies and virtual environments.

Memcord supports 28 MCP tools, fully compliant with the MCP spec dated 2025-03-26 and 2025-11-25. These tools include various extensions like tool annotations, result-size limits, and progress notifications to enhance AI interactions. On the summarization side, memcord supports four different summarizer backends: nltk, sumy, semantic, and transformers. This allows users to balance between speed and quality in summarization of conversational memory.

One of the clever design choices is the automatic project binding via .memcord files that live in project directories. This means memcord can auto-detect the current project context and activate the corresponding memory slot, streamlining developer workflows.

How the slot resolution priority system improves developer experience

The standout technical feature of memcord is its slot resolution priority system combined with automatic project binding:

  • When a memory slot is explicitly specified in a command or API call, memcord uses that slot.
  • If no explicit slot is provided, memcord falls back to the “active slot,” which is typically the last used or default slot.
  • If neither of the above is set, memcord tries to auto-activate a slot based on the presence of a .memcord file in the current working directory or its ancestors.

This layered fallback system means that developers rarely need to manually switch memory contexts during AI interactions. Instead, memcord intelligently selects the appropriate slot based on context, reducing friction and context-switching errors.

Under the hood, this slot mechanism is implemented in the core server logic, ensuring that all MCP tools and memory operations respect this priority order. The .memcord file acts as a lightweight project binding mechanism without requiring explicit configuration for each session.

The tradeoff here is the added complexity in slot management logic and the need for users to understand this priority system to troubleshoot context issues. However, in practice, it greatly enhances DX by eliminating repetitive context switching commands.

Quick start with memcord

Memcord provides straightforward installation and initial setup commands documented explicitly in the repository. The prerequisites are Python 3.10+ and the uv package manager, which the installer script handles automatically.

For macOS and Linux, you can install uv with:

curl -LsSf https://astral.sh/uv/install.sh | sh

For Windows PowerShell:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Once uv is installed, the quick start commands run platform-specific installer scripts:

macOS/Linux:

curl -fsSL https://github.com/ukkit/memcord/raw/main/install.sh | bash

Windows PowerShell:

irm https://github.com/ukkit/memcord/raw/main/install.ps1 | iex

This installer does the following:

  • Downloads and sets up memcord
  • Creates a Python virtual environment managed by uv
  • Generates MCP configuration files tailored for your platform
  • Configures integrations with Claude Desktop, Claude Code, VSCode, and Antigravity IDE

For manual installation or advanced usage, the repository also provides commands to clone the repo, create a virtual environment, install dependencies, and generate config files.

Who should consider memcord and final thoughts

Memcord is well-suited for developers and teams building AI applications that require persistent conversational memory without cloud dependencies. Its slot-based context isolation and automatic project binding simplify managing multiple AI workflows simultaneously.

The privacy-first approach is a clear advantage for use cases where sensitive data cannot leave the local environment. However, the tradeoff is that you must maintain the memcord server yourself, and it might not scale as easily as cloud-hosted solutions.

The codebase is well-structured with clear MCP spec compliance and multiple summarizer options to tune performance. For practitioners looking to embed a local AI memory server into their stack or experiment with MCP tooling, memcord provides a solid foundation.

In summary, memcord strikes a balance between developer ergonomics, privacy, and protocol compliance. Its slot resolution priority system is a neat pattern that reduces the cognitive load when juggling multiple AI contexts. If you are exploring self-hosted AI memory management or building AI agents with local context persistence, memcord is worth a close look.

→ GitHub Repo: ukkit/memcord ⭐ 65 · Python