Connect an agent
Give Claude Code — or any MCP client — access to your memory store.
nicemem speaks MCP — the protocol Claude Code, Claude Desktop, and most agent tools use to reach external services. One hosted endpoint, nothing to install:
https://nicemem.com/mcp
Transport is streamable HTTP; auth is your API key in the Authorization
header. Any MCP client that supports those two things can connect.
Create a key
In Settings → Keys, create a key and give it a name you'll recognize later ("laptop / Claude Code"). Two choices at creation:
- Write policy —
directcommits writes immediately;reviewsends them to your inbox for approval first. Usereviewfor agents you don't fully trust yet. - Space pin (optional) — pin the key to a space and it can only read that space plus open spaces, with writes landing in the pinned space by default. Good for scoping a work agent away from personal memories.
The key is shown once, at creation. Copy it then.
Connect Claude Code
claude mcp add --transport http nicemem https://nicemem.com/mcp \
--header "Authorization: Bearer nicemem_sk_…"
Other clients: point them at the endpoint above with the same header. No session setup, no OAuth dance — the key is the whole handshake.
The tools
Deliberately few:
| Tool | What it does |
|---|---|
add_memory |
Save a memory — content plus optional kind, title, tags, space. |
search_memories |
Semantic search, ranked by relevance × strength. |
recent_memories |
The timeline, newest first — orientation for agents. |
list_spaces |
The spaces this key can see. |
get_memory |
Fetch one memory by id. |
update_memory |
Edit a memory — corrections re-embed automatically. |
delete_memory |
Soft-delete a memory. |
Two things worth knowing: a save through a review key returns
status: "pending" — that's the inbox working, not an error. And writes
should pass a source_label (like claude-code) so you can tell your
agents apart in the timeline and inbox.
Teach it judgement
Connecting an agent gives it the tools; it doesn't teach it when to use them. That's what the agent prompt is for — install it in your agent's instructions and the difference in memory quality is immediate.