MCP-Memory

MCP-Memory is an open-source Model Context Protocol (MCP) server written in Python that equips AI agents with persistent, long-term memory across sessions. It stores memory records using the Open Knowledge Format (OKF v0.2) standard and indexes them in a local SQLite database with FTS5 full-text search for fast lookups, tag filtering, and keyword queries. The project is authored by Erfan Reed (fellowgeek) and released under the MIT License.

What It Is

MCP-Memory solves a core limitation of stateless AI agents: the inability to remember context, preferences, or progress between sessions. By acting as an MCP server, it exposes a set of tools that agents like Claude Desktop, Cursor, Windsurf, Antigravity, and Codex can call to store, retrieve, search, and delete structured memory snippets. Each memory is a self-describing OKF v0.2 Markdown document with YAML frontmatter, persisted both as a human-readable .md file on disk and as a row in a local SQLite index.

Dual-Layer Architecture

The server maintains two synchronized representations of every memory record:

• Human-Browseable OKF Directory: Every memory is written to a memory/ folder in the project root as a raw .md file. A hierarchical index.md provides progressive disclosure, and a log.md tracks update history.
• High-Performance SQLite Index: An FTS5-enabled SQLite database (.mcp_memory/memories.db) supports sub-20ms key lookups and instant keyword searches via automatic triggers.

This dual-layer design means developers can inspect and edit memories directly in their file system while agents benefit from fast programmatic access.

MCP Tools Exposed

The server registers six tools that agents can invoke over the MCP stdio transport:

• memory_store – Creates or updates a memory record with rich OKF metadata (type, key, namespace, tags, status, stale_after, sources, verified, generated_by).
• memory_retrieve – Fetches a specific memory by key and namespace.
• memory_search – Full-text and tag-filtered search across all stored memories.
• memory_get_last – Session-start directive: retrieves the last checkpoint so an agent knows where work was left off.
• memory_update_last – Milestone directive: updates the canonical session checkpoint whenever significant progress is made.

Setup and Deployment

MCP-Memory is designed for zero-boilerplate local deployment:

1. Clone the repository and run python3 setup.py — an interactive wizard auto-detects installed MCP clients (Antigravity, Claude, Cursor, Windsurf, Codex) and writes the correct configuration.
2. After setup, the AI client launches mcp-memory automatically via run.sh; no persistent server process needs to be managed manually.
3. Manual JSON or TOML configuration is also supported for clients that require it.
4. Storage paths are configurable via environment variables (MCP_MEMORY_PROJECT_ROOT, MCP_MEMORY_DB_PATH, MCP_MEMORY_DIR), enabling either per-project isolation or a single global memory store.

OKF Standard Compliance

Every memory record strictly follows the OKF v0.2 specification defined in the project's SPEC.md and OKF_RULES.md. Frontmatter fields include type, key, namespace, tags, generated, sources, verified, status, and stale_after. This standardization means memories are portable, inspectable, and interoperable with any tooling that understands the OKF format.

Current Status

The repository was created in August 2026 and had accumulated 171 stars and 10 forks within days of its initial push, according to the GitHub project metadata. The project is actively maintained on the main branch with 2 open issues at the time of indexing. The current memory format version is OKF v0.2, and the server self-identifies as mcp-memory/0.2.0 in generated metadata.