MCP support
mcp-memory-libravdb is the official MCP server for the libravdbd memory kernel. Give Claude Code, Codex, Cursor, Cline, Zed, or Aider persistent cognitive memory — semantic search, deontic gating, causal graph traversal, and workspace-aware multi-tenant isolation — through one protocol translation layer over gRPC. Zero memory intelligence reimplemented in the client.
brew install xDarkicex/homebrew-mcp-libravdb/mcp-memory-libravdbsudo apt install mcp-memory-libravdbgo install github.com/xDarkicex/mcp-libravdb-server/cmd/mcp-memory-libravdb@latestSearch across all collections with keyword, temporal, and cognitive kind filters. Results pass through obligation, prohibition, and permission gates before reaching the agent — unsafe recall never leaves the daemon.
Insert records with automatic cognitive kind classification — identity, constraint, decision, fact, preference, or episode. The daemon assigns the kind; the agent just writes.
Walk why-edges, how-edges, and associative links from any record through the daemon TopoRegistry. Trace a decision back to the constraint that shaped it, or forward to every fact it influenced.
The daemon pre-computes what the agent is likely to ask next from causal proximity and temporal recency. Context arrives before the turn executes — no extra round-trip, no prompt bloat.
Delete by ID with tenant-scoped authorization. Export and journal paths keep operator control over what stays and what goes. No orphaned memory, no hidden retention.
Node counts, cognitive kind breakdowns, tier distributions, daemon health, and tenant lists — all queryable through MCP tools with machine-readable JSON output.
By default, each project directory gets its own tenant key via CWD hashing — no configuration needed. Claude Code in one repo and Aider in another get separate memory namespaces automatically. Use --shared to unify tenants or --workspace for explicit naming.
Every MCP-compatible client connects the same way — one binary, one command. All six clients below use the identical stdio transport. The daemon does the intelligence; this server translates protocols.
Add via claude mcp add or a .mcp.json file in your project root. The tenant key auto-derives from the directory — every project gets isolated memory by default.
claude mcp add libravdb-memory -- mcp-memory-libravdb stdioConfigure in ~/.codex/config.toml under the mcp_servers section. Same stdio transport, same tenant isolation, same daemon backend.
mcp-memory-libravdb stdioDrop an MCP config at ~/.cursor/mcp.json. Cursor picks up the server on next launch — your agent gets cognitive memory without touching Cursor internals.
mcp-memory-libravdb stdioSettings → MCP Servers → add stdio. No config file needed — paste the command and Cline handles the rest. Per-project tenant scoping works the same as every other client.
mcp-memory-libravdb stdioAdd to ~/.config/zed/settings.json under context_servers. Zed's context protocol maps directly to MCP tools — memory surfaces in the assistant panel.
mcp-memory-libravdb stdioConfigure via .aider.conf.yml or the --mcp-server flag. Aider sees the same memory namespace as other clients in the same project directory.
aider --mcp-server libravdb-memory:stdio:mcp-memory-libravdbOne protocol translation layer. The daemon does the intelligence — this server translates MCP tool calls into gRPC requests over a Unix socket with HMAC authentication and per-tenant isolation.
Claude Code, Codex, and Aider in the same project all see each other's decisions, constraints, and architecture choices through the shared workspace causal graph. Use --shared to unify tenants.
OpenClaw or Hermes connect directly to the daemon with a chat tenant, getting read-only access to the code agent's workspace — answering questions about what was built with full context.
Each project directory gets its own tenant automatically via CWD hashing. No configuration needed. Move between projects and your memory scopes move with you.
mcp-memory-libravdb stdiomcp-memory-libravdb http --http-port 8082docker build -t mcp-memory-libravdb . && docker run mcp-memory-libravdb --version