@engram-ai-memory/mcp

MCP Server — connects Engram brain to any AI client (Claude Code, Cursor, Windsurf, Cline, and any MCP-compatible client). 21 tools for memory management, search, recall, decay, contradictions, tags, webhooks, reflection, and more.

Install

npm (manual)

npm install -g @engram-ai-memory/mcp

Installs two binaries: engram-mcp (the stdio server) and engram-store-session (writes a session summary, used by the Claude Code session-end hook).

Smithery (1-click — recommended for Claude Desktop)

smithery.ai/skills/ayvazyan10/engram — installs and configures automatically.

Desktop Extension (.mcpb)

Download engram-mcp.mcpb from GitHub Releases and open in Claude Desktop.

Setup

Add the block below to the config file your client reads — ~/.mcp.json is not one of them:

| Client | Config file | |---|---| | Claude Code (user scope) | ~/.claude.json | | Claude Code (project scope) | .mcp.json in the project directory | | Cursor | ~/.cursor/mcp.json | | Windsurf | ~/.codeium/windsurf/mcp_config.json |

engram setup --source <claude-code\|cursor\|windsurf> writes the right one for you.

{
  "mcpServers": {
    "engram": {
      "command": "npx",
      "args": ["-y", "@engram-ai-memory/mcp@latest"],
      "env": {
        "ENGRAM_DB_PATH": "~/.engram/engram.db",
        "ENGRAM_SOURCE": "claude-code"
      }
    }
  }
}

Set ENGRAM_SOURCE to your AI client: claude-code, cursor, windsurf, cline.

Restart your AI client. 21 tools appear automatically.

Tools

| Category | Tools | |---|---| | Memory | store_memory, search_memory, recall_context, add_knowledge, forget | | Stats | memory_stats, index_status, embedding_status | | Lifecycle | decay_sweep, decay_policy, re_embed | | Contradictions | check_contradictions, resolve_contradiction | | Tags | list_tags, tag_memory | | Webhooks | webhook_subscribe, webhook_list | | Plugins | plugin_list | | Reflection | request_reflection, store_reflection, get_reflections |

Recommended Workflow

Session start  → recall_context(task description)
During work    → store_memory(decisions, findings)
End of session → store_memory(session summary)

Environment Variables

| Variable | Default | Description | |---|---|---| | ENGRAM_DB_PATH | ~/.engram/engram.db | SQLite database path. A blank value counts as unset — not as the current directory, where a desktop host would put a database nobody can find | | ENGRAM_NAMESPACE_MODE | none (filter when ENGRAM_NAMESPACE is set) | none, filter, or isolated | | ENGRAM_NAMESPACE | — | Namespace used by filter/isolated modes | | ENGRAM_SOURCE | mcp-client | AI client identifier (e.g. claude-code, cursor, windsurf) | | ENGRAM_EMBEDDING_MODEL | Xenova/all-MiniLM-L6-v2 | Embedding model (must match the API server when sharing a database) | | ENGRAM_SYNC_URL | — | PostgreSQL connection string for multi-device sync. Unset = no sync engine is constructed | | ENGRAM_SYNC_MODE | auto | auto, manual, or off | | ENGRAM_SYNC_INTERVAL | — | Background sync interval, in milliseconds | | ENGRAM_SYNC_ENCRYPTION_KEY | — | Passphrase for end-to-end encryption of synced rows |

The four sync variables are validated at startup: a bad mode or interval stops the server with a message naming the variable, rather than becoming a timer that never stops. See Cloud Sync.

Links

• GitHub
• MCP Tools Reference
• Website

License

MIT