nlqdb

Solve · Agent builders

Is there an MCP server that gives my AI agent memory?

If you want an MCP server that gives your AI agent memory, point your coding agent at nlqdb's hosted one with one command — `claude mcp add --transport http nlqdb https://mcp.nlqdb.com/mcp`. Unlike a blob-store memory server, it's a real Postgres your agent writes to and queries in English, so 'how many' and 'top N' are SQL, not guesses.

Search 'agent memory MCP server' and you land on a pile of registry entries — most are a thin wrapper over a JSON file or a vector store, exposing `store_memory` / `get_memory` tools that write blobs and return the top-k similar ones. That covers recall. It falls apart the moment you want to count what the agent remembered, group it by user, or see what changed this week — a blob has no grain and a vector store has no query planner, so those answers become the LLM doing arithmetic over a list.

The snippet that solves it.

> count of remembered items grouped by kind, most first

What nlqdb does for this

  • One command — `claude mcp add --transport http nlqdb https://mcp.nlqdb.com/mcp` — points Claude Code, Cursor, or Codex at nlqdb's hosted MCP server.
  • The `nlqdb_query` tool provisions a real Postgres from the agent's first English goal — no connection string, no schema, no separate create call.
  • Because memory is typed rows, `GROUP BY`, top-N, and per-period rollups run as SQL — the aggregates a blob or vector-store memory server can't.
  • Every answer returns the rows plus the compiled SQL, and tenant isolation is a fail-closed RLS policy in the engine, not a `WHERE` you remember.

Drop into any HTML page

<nlq-data goal="count of remembered items grouped by kind, most first"></nlq-data>

The question a blob-store memory server can't answer — group what the agent remembered by kind and count each — is one GROUP BY when memory is typed rows in a real database.

What this replaces

What nlqdb doesn't try to do here

  • It's a hosted server (`mcp.nlqdb.com/mcp`), not a self-hostable local MCP binary you run beside your agent — the FSL-1.1 self-host container is on the roadmap, not shippable today.
  • The dedicated `nlqdb_remember` verb and the opinionated `agent_memory_v1` schema (facts/episodes/entities) are authed and gated (`MEMORY_PRESET` is dark, `SK-PIVOT-010`); today the live path is `nlqdb_query` provisioning and querying a database, and it rejects anonymous writes.
  • No native vector search — nlqdb is Postgres-first; if what you need from a memory server is unstructured similarity recall over chat-text, that's Mem0 or pgvector's job, and nlqdb composes with it.

Questions buyers ask

Is there an MCP server that gives my AI agent memory?
Yes — nlqdb's hosted MCP server. Run one command (`claude mcp add --transport http nlqdb https://mcp.nlqdb.com/mcp`) and your coding agent gets a real hosted Postgres via the `nlqdb_query` tool: it provisions the database from the first English goal, then reads and writes it in English with the SQL shown. It's memory you can aggregate over, not just a blob store.
How is this different from a memory MCP server that stores and retrieves memories?
Most 'memory' MCP servers wrap a JSON file or a vector store: `store_memory` writes a blob, `get_memory` returns the top-k similar ones. That's recall. nlqdb stores memory as typed rows in Postgres, so on top of recall you can `GROUP BY`, count, and roll up — 'the 5 things this agent remembers most' is a query, not the LLM adding up search hits.
Which coding agents and MCP hosts can install the nlqdb memory server?
Any host that speaks MCP. Cursor and VS Code install from a one-click link, Claude Code from one command, and Codex from a config-file line (all sourced from the host descriptors in `mcp-install.ts`); Claude Desktop, Windsurf, and Zed connect via their custom-connector settings. It's a hosted remote server at `mcp.nlqdb.com/mcp` — no local binary to run.
Do I get dedicated remember and recall tools, or just query?
Today the live tool is `nlqdb_query`, which both provisions the database and answers over it in English. A dedicated `nlqdb_remember` verb and the opinionated `agent_memory_v1` schema (facts, episodes, entities with per-agent isolation and TTL) exist but are authed and gated behind `MEMORY_PRESET` (`SK-PIVOT-010`) — coming, not anonymous today. Until then the agent provisions and queries its memory tables through `nlqdb_query`.

Where this pain shows up in public

Enduring discussion hubs where you can verify the theme without taking our word for it. We don't quote individual posts; we cite search-result and subreddit URLs that stay live as new threads land.

Try nlqdb in 30 seconds

No sign-in. The anonymous database lasts 72 hours; adopt it with one click if you keep it.

Start with a goal →

Looking at this from a different angle? Browse all solve pages or browse competitor comparisons.