diff --git a/graph/configurations/cognitive-memory-mcp-read-only-deployed-to-discord-coordinat-5c4a03.md b/graph/configurations/cognitive-memory-mcp-read-only-deployed-to-discord-coordinat-5c4a03.md
new file mode 100644
index 00000000000..877e0901af2
--- /dev/null
+++ b/graph/configurations/cognitive-memory-mcp-read-only-deployed-to-discord-coordinat-5c4a03.md
@@ -0,0 +1,47 @@
+---
+id: 5c4a03f4-0c2c-4ef4-9c64-29cf9211be4a
+type: configuration
+title: "Cognitive Memory MCP (read-only) deployed to Discord Coordinator LXC 301"
+tags: [cognitive-memory, discord-coordinator, mcp, homelab, lxc-301, configuration]
+importance: 0.8
+confidence: 0.8
+created: "2026-02-21T05:24:06.417617+00:00"
+updated: "2026-02-21T05:24:06.417617+00:00"
+---
+
+# Cognitive Memory MCP Read-Only Deployment on LXC 301
+
+## Context
+
+Deployed a read-only instance of the cognitive memory MCP server to LXC 301 (ai-pm / discord-coordinator) so the Discord bot agent can query memories without being able to write or modify them.
+
+## Files Deployed
+
+All files placed in `/home/discord-bot/.claude/skills/cognitive-memory/`:
+- `client.py` - core memory client
+- `mcp_server.py` - full MCP server (not used directly)
+- `SKILL.md` - skill documentation
+- `mcp_server_readonly.py` - new readonly wrapper (key addition)
+
+## Key Configuration
+
+- Symlink: `~/.claude/memory` -> `memory-sync/` (git-synced memory store)
+- MCP config: `~/.claude/.mcp.json` points to `mcp_server_readonly.py`
+- `settings.json`: `ENABLE_TOOL_SEARCH=true` for lazy-loading of MCP tools
+
+## Readonly Wrapper Design
+
+`mcp_server_readonly.py` exposes **11 read-only tools**:
+- `memory_recall`, `memory_get`, `memory_search`, `memory_related`
+- `memory_edge_get`, `memory_edge_search`, `memory_reflection`
+- `memory_stats`, `memory_core`, `memory_tags_list`, `memory_tags_related`
+
+**7 write tools are blocked** with "Memory is read-only on this server" error:
+- `memory_store`, `memory_relate`, `memory_embed`, `memory_decay`
+- `memory_episode`, `memory_config`, `memory_reflect`
+
+## Notes
+
+- No Ollama required — semantic search gracefully falls back to keyword-only search
+- 538 memories indexed on initial deploy
+- Requires `client.py reindex` on first deploy to build `_index.json` from scratch