A persistent memory MCP server with FTS5 fuzzy search. Compatible with Bun and Node.js.
- Persistent Storage: Data stored in
~/.mcp-memory/memory.dbsurvives npx/bunx cache clears - Fuzzy Search: FTS5 with trigram tokenization for substring matching (
pyth→python) - Token Efficient: Results in TOON format for minimal token usage
- Auto Maintenance: Prunes entries older than 30 days, creates backups on startup
- Cross Runtime: Works with both Bun and Node.js
npx @renzynx/memory-mcp
# or
bunx @renzynx/memory-mcpAdd to ~/.config/opencode/opencode.jsonc:
Add to claude_desktop_config.json:
{
"mcpServers": {
"memory": {
"command": "npx",
"args": ["@renzynx/memory-mcp"]
}
}
}Store information with a category.
save_memory(content: string, category: string)
Categories: preferences, facts, context, projects, conventions
Fuzzy search stored memories. Returns TOON format or Ø if empty.
search_memories(query: string)
List all unique categories. Returns TOON format or Ø if empty.
list_categories()
Add to your agent system prompt:
## Memory System
You have access to a persistent memory system via MCP tools. Use it proactively to remember important information across sessions.
### Tools Available
- `save_memory(content, category)` - Store information with a category
- `search_memories(query)` - Fuzzy search (supports partial matches like "pyth" → "python")
- `list_categories()` - View all memory categories
### When to Save Memories
- User preferences (coding style, tools, frameworks, communication preferences)
- Project context (architecture decisions, file structures, conventions)
- Facts about the user (name, role, team, timezone)
- Recurring tasks or workflows
- Corrections or clarifications the user provides
- Important decisions and their rationale
### Categories to Use
- `preferences` - User preferences and settings
- `facts` - Information about the user or their environment
- `projects` - Project-specific context and decisions
- `conventions` - Coding standards and patterns
- `context` - Session or task context worth preserving
### When to Search Memories
- At the start of conversations to recall user context
- Before making assumptions about preferences
- When the user references something previously discussed
- Before suggesting tools, patterns, or approaches
### Output Format
Results return in TOON format for token efficiency. "Ø" means no results found.
### Best Practices
- Save incrementally, not everything at once
- Use specific, searchable content
- Search before asking the user to repeat themselves
- Update memories when information changes (save new version)
MIT
{ "mcp": { "memory": { "type": "local", "command": ["bunx", "@renzynx/memory-mcp"] } } }