@agent-memory/mcp
v0.2.0
Published
MCP server for agent-memory: long-term, cross-project memory system for AI agents
Readme
@agent-memory/mcp
MCP server for agent-memory — long-term, cross-project memory for AI agents.
For AI agents that don't have shell access (e.g. Claude Desktop, Windsurf, Cline), this package exposes agent-memory as an MCP server over stdio.
Install
npm install -g @agent-memory/mcpSetup
Add to your MCP client config:
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"agent-memory": {
"command": "agmem-mcp"
}
}
}Claude Code (~/.claude/settings.json):
{
"mcpServers": {
"agent-memory": {
"command": "agmem-mcp"
}
}
}Cursor (.cursor/mcp.json):
{
"mcpServers": {
"agent-memory": {
"command": "agmem-mcp"
}
}
}Tools
14 tools organized in 4 groups:
Memory CRUD
| Tool | Description |
|------|-------------|
| memory_add | Store a new memory with optional tags and digest |
| memory_get | Retrieve one or more memories by ID |
| memory_update | Update a memory's content, tags, or digest |
| memory_delete | Permanently delete memories by ID |
Search & Browse
| Tool | Description |
|------|-------------|
| memory_search | Hybrid search (full-text + vector + RRF fusion) |
| memory_list | Browse and filter memories with pagination |
| memory_tags | List all tags with usage counts |
| memory_stats | Show memory store statistics |
Source Management
| Tool | Description |
|------|-------------|
| source_list | List configured memory sources |
| source_run | Run source commands (embed, rebuild, status) |
Config
| Tool | Description |
|------|-------------|
| config_set | Set a configuration value |
| config_get | Get a configuration value |
| config_list | List all configuration values |
| config_delete | Delete a configuration value |
First Run
On first use, initialize the embedding model for semantic search:
memory_add → works immediately (FTS search available)
source_run { command: "init" } → downloads embedding model (~65MB) for vector searchWithout initialization, search falls back to keyword-only (FTS5). Semantic search activates once the model is downloaded.
License
MIT
