Memclaw

Name: Memclaw
Author: caura-ai

@caura-ai

About Memclaw

MemClaw — persistent memory for AI agent fleets (OSS)

#mcp #knowledge-graph #openai #memory-management #multi-agent-systems #ai-agents #claude #rag #fastapi #pgvector #llm-memory #agentic-ai #ai-infrastructure #mcp-server #agent-memory

Basic information

Transports

stdio

Publisher

caura-ai

Submitted by

Jyolsna

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "memclaw": {
      "url": "http://localhost:8000/mcp",
      "headers": {
        "X-API-Key": "standalone"
      }
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Memclaw?

Memclaw is open-source fleet memory for multi-tenant, multi-agent AI fleets. Agents store what they learn, recall what the fleet knows, and compound knowledge across interactions instead of repeating mistakes. It is built for production deployments with dozens to thousands of agents sharing governed memory.

How to use Memclaw?

Clone the repository, configure a .env file (e.g., with an OpenAI API key), and run docker compose up -d to start Postgres + pgvector + Redis + the API. Write a memory via curl to /api/v1/memories and search with /api/v1/search. Alternatively, connect an MCP client using the server URL https://memclaw.net/mcp with an API key from the dashboard. Three deployment paths are offered: managed platform, self-hosted Docker, and OpenClaw plugin.

Key features of Memclaw

Multi‑tenant, multi‑agent governed memory.
Agents write plain text; enriched with LLM‑inferred fields.
Cross‑agent outcome propagation and fleet‑wide trust tiers.
Open‑source, self‑hosted or managed platform.
MCP and REST API with scoped credentials.

Use cases of Memclaw

Fleet of customer‑support agents sharing resolved solutions.
Internal knowledge base for AI agents on company policy and APIs.
Team of coding agents learning from each other’s debugging patterns.
Multi‑agent systems requiring governed, compounding memory.

FAQ from Memclaw

What distinguishes Memclaw from single‑agent memory systems?

Memclaw is architected for fleets: scoped memory, cross‑agent outcome propagation, and fleet‑wide trust tiers, not a single long conversation. It competes on latency, token efficiency, and governance at scale.

What are the runtime dependencies?

PostgreSQL with pgvector for vector storage, Redis for the event bus, and optionally an AI provider (OpenAI, Gemini, Anthropic, OpenRouter) or a self‑hosted embedder (BAAI/bge-m3) for embeddings and enrichment.

Where does my data live?

In self‑hosted mode, data is stored in your local PostgreSQL instance. The managed platform (memclaw.net) hosts the database and infrastructure for you.

What transports and authentication are supported?

REST API and MCP transport. Auth uses tenant‑scoped or agent‑scoped API keys (prefix mc_). Standalone mode bypasses auth with X-API-Key: standalone.

Are there any known limits or performance benchmarks?

Without an AI key, dummy embeddings are used (no semantic search). In production at eToro: 300+ agents, 26,500+ memories, 1,372 shared skills, 23 ms p50 search.

Comments

88.1K