Build a Persistent, Searchable Agent Memory
MCP server providing persistent, searchable memory for AI agents, with search, write, read, and stats tools plus a usage dashboard.
Why it matters
Establish a long-term memory for your AI agent that automatically syncs with project documentation, enabling efficient knowledge retrieval and management.
Outcomes
What it gets done
Create a searchable memory bank for AI agents.
Automatically sync agent memories with project documentation.
Read, write, and search long-term memories via MCP tools.
Visualize memory usage with a dedicated dashboard.
Install
Add it to your toolbox
Run in your project directory:
curl -fsSL https://spark.entire.vc/get/ag-agent-memory-mcp | bash Overview
Agent Memory Skill
Agent Memory MCP Server is a self-hosted memory bank for AI agents, exposing search, write, read, and stats tools over MCP plus a local dashboard. It syncs with project documentation so agents can persist architecture decisions and patterns across sessions instead of starting from scratch. Reach for it when an MCP-capable agent needs memory that survives between sessions and you are comfortable running a local Node.js server and dashboard.
What it does
Agent Memory is an MCP server that gives AI agents a persistent, searchable memory bank, automatically syncing with project documentation. It stores architecture decisions and patterns so agents can search, write, read, and inspect what they have learned across sessions. Running locally as an MCP server, it exposes four tools: memory_search for finding memories by query, type, or tags, memory_write for recording new knowledge with a key, type, content, and optional tags, memory_read for retrieving a specific memory by key, and memory_stats for viewing usage analytics. A standalone dashboard is also included for visualizing memory usage.
When to use - and when NOT to
Use this skill when an agent needs to retain and retrieve knowledge - architecture choices, patterns, past decisions - across a project over time instead of re-deriving it every session. It fits workflows where an MCP-capable agent already talks to local tooling and can run an additional Node.js server. It is not a fit for environments without Node.js v18+ or where running a local MCP server plus a dashboard process is impractical.
Inputs and outputs
Setup requires Node.js v18+, cloning the agentMemory project into the agent's workspace:
git clone https://github.com/webzler/agentMemory.git .agent/skills/agent-memory
followed by npm install and npm run compile. The server starts with npm run start-server <project_id> <absolute_path_to_target_workspace>. Each memory is written and read via a key; memory_write also takes type, content, and optional tags, while memory_search accepts a query plus optional type and tags filters and returns matching memories. memory_stats returns usage analytics with no arguments required.
Integrations
The dashboard runs as a separate process started with npm run start-dashboard <absolute_path_to_target_workspace> and is viewed in a browser at http://localhost:3333.
Who it's for
Teams running AI coding agents that need memory to persist across sessions rather than starting from a blank context each time, and who want a lightweight, self-hosted MCP memory server they can inspect through a local dashboard.
FAQ
Common questions
Discussion
Questions & comments · 0
Sign In Sign in to leave a comment.