Long-term memory in Cursor
Cursor connects to khive's hosted MCP endpoint with an API key. That connection gives Cursor access to memory tools that store and find memories in your khive workspace.
Why Cursor forgets between chats
Cursor's documentation says it in one line: “Large language models don't retain memory between completions.” What carries from one chat to the next is what Cursor puts at the start of the model context, and that is your rules. Project rules live in .cursor/rules in the repository, user rules apply across your Cursor environment, and an AGENTS.md at the project root works in place of project rules. Cursor documents all three in Rules.
Use them. Three cases fall outside them:
- Something that came up in a chat. A rule is a file someone writes. Why a migration was rolled back, or which flag the staging server needs, is in no rule until a person puts it there.
- A long list of facts. Cursor's guidance is to keep rules under 500 lines, and a rule set to always apply is included in every chat session. A list of decisions that grows every week does not fit that.
- A second tool.
.cursor/rulesand user rules are read by Cursor. A memory that Claude Code, Codex or ChatGPT should also see, in any repository, has to live somewhere all of them can reach.
Where khive fits, and where it does not
khive keeps memories in a hosted workspace instead of a file in one repository. Cursor saves one with remember and finds it in a later chat with recall_memories, which searches by meaning and returns the closest matches, so only those matches enter the context window, not the whole store. Every client connected to the same khive account reaches the same memories: Cursor on another machine, Claude Code, Codex or ChatGPT.
The agent decides when to call a tool, so to be sure something is kept, say so (“Save to khive that staging needs the TLS flag”). A rule that always applies can ask for the same thing in every chat. khive stores what the agent saves. It does not copy your chats, and it does not replace your rules.
What khive gives you
The remember tool stores a memory, and recall_memories finds memories by meaning.
Every client is offered the same tools, including the typed knowledge graph, tasks, and reference knowledge.
How to set it up
Create a khive account and plan, create an API key, add it to an environment variable, and configure khive in Cursor's mcp.json.
Cursor can load the configuration for all projects or for one project; approve the server with the Cursor CLI. Follow the Cursor connection steps and the Quickstart for details.
What it costs
| Plan | Monthly | Annual | Trial |
|---|---|---|---|
| Start | $8 | $80 | none |
| Plus | $25 | $250 | 14 days, on monthly |
| Pro | $100 | $1,000 | none |
Prices are in US dollars. There is no free tier. Plus monthly starts with a 14-day free trial. See Plans for plan details.
Limits and what it does not do
Cursor's own MCP sign-in needs dynamic client registration, which khive does not offer, so Cursor connects with an API key.
Plans set request and stored-record limits, and the dashboard shows remaining included usage as a percentage.