@nlqdb/mcp

Analytical memory for AI agents, over MCP — a real Postgres your agent can GROUP BY / JOIN / aggregate over in natural language, not just a recall store. This package is the local stdio transport: it reads an API key from the environment, so it works headless (CI, Docker, air-gapped, coding agents that can't open a browser).

Don't want a local process? Point your host at the hosted server https://mcp.nlqdb.com/mcp instead — OAuth in the browser, nothing to install.

Install

Create an sk_mcp_… MCP key at app.nlqdb.com/app/keys, then add this to your MCP host's config:

{
  "mcpServers": {
    "nlqdb": {
      "command": "npx",
      "args": ["-y", "@nlqdb/mcp"],
      "env": { "NLQDB_API_KEY": "sk_mcp_…" }
    }
  }
}

sk_mcp_… is bound to one host + device and reaches every tool here except nlqdb_connect_database (attaching an external database stays an account action); revoke it on its own at /app/keys. An sk_live_… account key also works if you need that one tool too. A pk_live_… key is pinned to one database and can only call nlqdb_query.

Drop-in skill — teach the agent when to remember

Connecting the server gives your agent the tools. It doesn't tell it what is worth remembering, or that it can aggregate over what it stored. One command installs a skill that does, into a repo your agent already reads:

npx skills add https://github.com/nlqdb/nlqdb/tree/main/apps/web/public/agent-artifacts/nlqdb-memory

Run against the live skills CLI 2026-07-27 — no account, no publish. It writes .agents/skills/nlqdb-memory/SKILL.md (the cross-agent directory Cursor and Codex read directly), a .claude/skills/nlqdb-memory symlink for Claude Code, and a skills-lock.json. It does not write a .cursor/rules/ file and does not edit AGENTS.md — a host that reads only AGENTS.md needs AGENTS.snippet.md appended by hand. All four host artifacts: nlqdb.com/agent-artifacts.

Tools

| Tool | Does | |---|---| | nlqdb_query | Ask in plain English; returns rows plus the compiled SQL. The database materialises on first reference — there is no create tool. | | nlqdb_remember | Write a typed memory row the agent can query later. | | nlqdb_list_databases | List your databases. | | nlqdb_describe | Describe a database's schema. | | nlqdb_connect_database | Attach your own Postgres or ClickHouse. |

Destructive plans (INSERT/UPDATE/DELETE/DDL) come back as requires_confirm: true with a diff — re-call with confirm: true to commit. NLQDB_MCP_DEBUG=1 prints stack traces on fatal errors.

More

Per-host config blocks and one-click installs: docs.nlqdb.com/mcp · what agent memory buys you · design + contributing