RELATED 01
agentmemory
Gives coding agents a persistent memory they can save to, search and share across sessions
30kGitHub stars
19k/wknpm downloads
78popularity score
AI DevKit Memory
Local SQLite memory for coding agents to save and look up project rules, conventions and fixes
By Hoang Nguyen (individual) TypeScript Local · stdio Knowledge and memoryDeveloper tools
48 popularity score ★ 1.7k ↓ 381/wk
Connect your client
Local (npm)
Add in one click
Opens the app, which asks before it adds the server.
Or add it by command or config
Claude CodeTerminal
claude mcp add ai-devkit-memory -- npx -y @ai-devkit/memoryClaude Desktopclaude_desktop_config.json
{
"mcpServers": {
"ai-devkit-memory": {
"command": "npx",
"args": [
"-y",
"@ai-devkit/memory"
]
}
}
}Cursor~/.cursor/mcp.json
{
"mcpServers": {
"ai-devkit-memory": {
"command": "npx",
"args": [
"-y",
"@ai-devkit/memory"
]
}
}
}VS Code.vscode/mcp.json
{
"servers": {
"ai-devkit-memory": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@ai-devkit/memory"
]
}
}
}Codex~/.codex/config.toml
[mcp_servers.ai-devkit-memory]
command = "npx"
args = ["-y", "@ai-devkit/memory"]Gemini CLITerminal
gemini mcp add --scope user ai-devkit-memory npx -- -y @ai-devkit/memoryWhat it does
AI DevKit Memory is the MCP server inside the AI DevKit project, published on npm as @ai-devkit/memory. It keeps a small knowledge base for coding agents in a SQLite file on your own disk, so that a decision made in one session (for example, how API responses are shaped) can be found again in the next session instead of being pasted into every prompt.
The server has three tools: one stores a titled entry with optional tags and a scope (global, a named project or a named repo), one edits an entry by its id, and one searches. Search uses SQLite FTS5 full-text matching with BM25 ranking, boosts results that share the tags you pass, and returns 5 results by default, up to 20 with `limit`. Duplicate content is rejected on store.
An optional semantic mode, switched on with "memory": {"semantic": true} in the project's .ai-devkit.json, adds local MiniLM embeddings and fuses them with the full-text ranks. The model (about 23 MB) is downloaded once to ~/.ai-devkit/models and then runs offline; if it is missing, search falls back to full-text results.
By the code at memory@0.20.3, the database lives at ~/.ai-devkit/memory.db, is opened in WAL mode, and the server still accepts the older dotted tool names (memory.storeKnowledge and so on) for agents with stale prompts. The wider AI DevKit CLI (`ai-devkit init`) writes this server into an agent's config for you, but the server can also be added to any MCP client by hand.
What it needs
| Runtime | Node.js 20+ |
|---|
Node.js 20.20.0 or newer (from the package's engines field) with npx. It runs over stdio on your machine and needs no account or key. The better-sqlite3 dependency is a native module, so the install must be able to fetch or build it for your platform. Semantic mode needs a one-time download of the MiniLM model, done on first use or with `ai-devkit memory semantic download` from the separate ai-devkit CLI.
Limits
There is no delete tool: entries can be stored, edited and searched, but removing one is not offered through MCP. Titles must be 10-100 characters, content 50-5000 characters, tags at most 10, and search queries 3-500 characters. Search returns at most 20 results per call. Semantic search is off by default and is configured per project in .ai-devkit.json, not through an environment variable. The npm package is MIT-licensed while the repository as a whole is Apache-2.0. The docs' client snippet omits -y for npx, which can leave npx waiting for an install prompt; the command here adds it.
Who it suits
Suits a developer who wants coding agents (Claude Code, Cursor or others) to keep a shared, local record of project conventions and past fixes without a hosted service or sign-in. It writes only to its own SQLite file in your home directory. Pick something else if you need memory shared across machines or a team, a general knowledge graph, or a way for the agent to delete entries; and if you already use the full AI DevKit CLI, `ai-devkit init` wires this server in for you.
Tools 3
memory_storeKnowledge- Save a new entry with a title, markdown content, optional tags and a global, project or repo scope.
memory_updateKnowledge- Change the title, content, tags or scope of an existing entry by its id.
memory_searchKnowledge- Find stored entries matching a task description, with tag boosting, scope filter and a result limit (default 5, max 20).