One memory store. Every MCP client.
Remnic speaks MCP natively over both stdio and HTTP. Any client that supports the protocol — Cursor, Cline, Windsurf, Amp, Replit, or your own custom agent — can recall, store, and search memories through the same set of tools.
Cursor
Start the Remnic server and point Cursor's MCP config at the HTTP endpoint.
export REMNIC_AUTH_TOKEN="$(openssl rand -base64 32)"
npx remnic-server --host 127.0.0.1 --port 4318 \
--auth-token "$REMNIC_AUTH_TOKEN" Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"remnic": {
"url": "http://localhost:4318/mcp",
"headers": {
"Authorization": "Bearer ${REMNIC_AUTH_TOKEN}"
}
}
}
}
Restart Cursor. All 44 Remnic MCP tools become available, including
remnic.recall, remnic.memory_store,
remnic.entity_get, and the LCM expansion tools.
Replit
Replit only supports MCP — no hooks — so recall is on-demand rather than automatic.
remnic connectors install replit The connector generates a token and shows the exact config block to paste into Replit's Integrations pane.
Cline, Windsurf, Amp, and generic MCP clients
Any MCP-compatible client takes the same HTTP config as Cursor — point
it at http://localhost:4318/mcp with the bearer token.
For stdio-based clients, use:
openclaw engram access mcp-serve # OpenClaw-hosted stdio path
# or, standalone:
npx remnic-server --stdio --auth-token "$REMNIC_AUTH_TOKEN" Full per-client setup snippets live in the Connector Setup Guide.
HTTP REST for custom agents
If you're building a custom agent framework, skip MCP and hit the REST API directly.
curl -X POST http://localhost:4318/engram/v1/recall \
-H "Authorization: Bearer $REMNIC_AUTH_TOKEN" \
-H "Content-Type: application/json" \
-d '{"query":"authentication patterns","topK":8,"mode":"minimal"}'
Full API reference:
docs/api.md.
Routes under /engram/v1/... during the v1.x compatibility
window.
Multi-tenant deployments
Run Remnic as a standalone server that multiple agents and users share. Isolate tenants with namespace policies, multi-token auth, and per-namespace principals. Ideal for teams and multi-machine setups.
See the Standalone Server Guide and Deployment Topologies doc for localhost, LAN, remote, containerized, and stand-alone setups.
MCP tools reference
| Tool | Purpose |
|---|---|
engram.recall | Retrieve relevant memories for a query. |
engram.recall_explain | Debug the last recall — per-tier breakdown. |
engram.memory_get | Fetch a specific memory by ID. |
engram.memory_store | Store a new memory. |
engram.memory_timeline | View a memory's lifecycle history. |
engram.entity_get | Look up a known entity. |
engram.day_summary | Generate structured end-of-day summary. |
engram.observe | Feed conversation messages into the pipeline. |
engram.lcm_search | Full-text search across archived conversation history. |
engram_context_expand | Retrieve raw lossless messages for a turn range. |
engram.review_queue_list | View the governance review queue. |
engram.suggestion_submit | Queue a memory for review. |