apex-memory is an MCP server that gives AI coding agents persistent memory across sessions. Save decisions, bugs, patterns, and context — recall them exactly when needed.
Not a general-purpose memory layer bolted on after the fact. Every design decision is optimized for the coding agent workflow.
claude mcp add apex-memory -- npx apex-memory
Works immediately. No configuration, no Docker, no accounts.
All data lives in ~/.apex-memory/memory.db — a local SQLite database with WAL mode. Nothing leaves your machine.
FTS5-powered search across all memories. Supports AND, OR, NOT, and phrase queries. Find anything instantly.
Categories designed for real dev workflows: decision, architecture, bug, pattern, preference.
Start and end coding sessions. Group related memories together. Review what the agent accomplished across sessions.
Built on the Model Context Protocol from day one — not a wrapper. Works with any MCP-compatible editor, today and tomorrow.
apex-memory adds 9 tools to your agent's toolkit, covering the full memory lifecycle — including importance scoring, access tracking, and auto-decay (v0.2).
| Tool | What it does |
|---|---|
| save_memory | Save knowledge, decisions, patterns, or context with categories, tags, and project |
| search_memories | Full-text search across all memories — supports AND, OR, NOT, phrases (FTS5) |
| recall | Retrieve recent memories filtered by project, category, or tags |
| delete_memory | Remove a specific memory by ID |
| start_session | Begin a coding session tied to a project — returns a session ID |
| end_session | End a session with a summary of what was accomplished |
| list_sessions | View past coding sessions, optionally filtered by project |
| decay_memories | Archive old low-importance memories that were never accessed — keeps the store clean automatically |
| memory_stats | Get statistics about stored memories, sessions, and projects |
// example: agent saves a decision, recalls it next session
Those are great general-purpose memory systems. apex-memory is purpose-built for coding agents.
| apex-memory | Mem0 | Letta | OneContext | |
|---|---|---|---|---|
| Built for coding agents | Yes | General purpose | General purpose | General purpose |
| MCP-native | Yes | No | No | No |
| Zero infrastructure | Yes (SQLite) | Requires server | Requires server | Cloud-only |
| Install time | 30 seconds | Minutes | Minutes | Minutes |
| Works with Claude Code | Native | Manual | Manual | No |
| Session tracking | Built-in | No | No | No |
| Search method | FTS5 (exact) | Vector | Vector | Vector |
| Open source | MIT | Partial | Yes | No |
apex-memory is MCP-native — it connects to any editor that supports the Model Context Protocol over stdio.
One command: claude mcp add apex-memory -- npx apex-memory. Native MCP support, zero config.
Add to .cursor/mcp.json. Works with Cursor's AI agent out of the box. Setup guide →
Add to ~/.codeium/windsurf/mcp_config.json. Works with Cascade. Setup guide →
If it supports MCP stdio transport, it works. Point it at npx apex-memory. Generic guide →
The free tier is fully functional — unlimited sessions, unlimited memories, forever.
One command. Zero infrastructure. Works today.