Show HN: MCP Memory – Fast Agent Memory Using Google's OKF and SQLite FTS5
MCP Memory is a new tool that provides AI agents with persistent, searchable long-term memory, leveraging Google's Open Knowledge Format (OKF) and SQLite FTS5 for efficient indexing. It aims to solve the critical problem of agents retaining context across sessions, a frequent pain point in AI development. The project sparked discussion on its technical choices, performance, and how it differentiates itself from existing agent memory solutions.
The Lowdown
MCP Memory is an innovative Model Context Protocol (MCP) server designed to equip AI agents with robust, persistent, and searchable long-term memory capabilities. It stores memory records using the Open Knowledge Format (OKF v0.2) and indexes them with a local SQLite instance, complete with FTS5 for full-text search, enabling fast lookups and content retrieval.
Key features include:
- Persistent State Across Sessions: Agents can store, retrieve, update, and delete memory snippets that persist beyond individual chat turns or sessions.
- OKF Standard Compliance: All memory items are formatted as OKF v0.2 Markdown documents with structured YAML frontmatter, ensuring standardized data representation.
- Dual-Layer Architecture: Combines human-readable Markdown files on disk for easy browsing with a high-performance SQLite database for rapid indexing and search.
- Namespace Isolation: Allows for contextual separation of memories, such as user preferences or project-specific architecture details.
- Zero Boilerplate Setup: Features an interactive setup wizard for easy integration with popular AI tools like Claude Desktop, Cursor, and Antigravity.
- Comprehensive MCP Tools: Exposes primary tools like
memory_store,memory_retrieve,memory_search,memory_get_last(for session checkpoints), andmemory_update_lastfor agent interaction.
By offering a structured yet flexible approach to memory management, MCP Memory aims to enhance the intelligence and continuity of AI agent operations, moving beyond transient context windows to a more enduring knowledge base.
The Gossip
Memory Maze Musings
Many commenters questioned the novelty of MCP Memory, asking how it differs from numerous existing agent memory solutions or simpler approaches like grepping Markdown files. The discussion touched on whether the complexity of such systems truly benefits agents, with some users noting that 'MCP things... tend to slow down the agent and waste tokens more than they end up helping.' The author and others pointed out advantages like portability across different MCP-enabled AI platforms and the structured nature of OKF over plain text files, arguing that a dedicated system offers more than basic file search.
OKF's Operational Oomph and SQLite's Swiftness
The technical choices behind MCP Memory, particularly the use of Google's Open Knowledge Format (OKF) and SQLite FTS5, were a focal point. Commenters asked for clarification on why OKF was preferred over plain Markdown and if it offered tangible benefits. The author clarified that OKF is essentially Markdown with frontmatter for metadata, enabling structured storage. The use of SQLite FTS5 for 'fast agent memory' was praised as a 'pragmatic architectural choice,' though some users requested benchmarks to prove performance gains over built-in agent memory systems or simpler methods.
Agent Application & Adaptation
Discussion also explored the broader implications of agent memory systems. Questions arose regarding the model-dependency of agent tool use—whether different LLMs require specific prompting to effectively utilize external memory. Concerns were voiced about the 'staleness' of recorded facts in constantly changing codebases, prompting a comparison to personal note-taking where information can still be valuable despite needing periodic review. One commenter also inquired about how this system contrasts with knowledge graph implementations, particularly in managing ontologies and leveraging traversals.