GETTING STARTED

Quick Start

From zero to memory + code search in five minutes. One page — no docs site to get lost in.

01Create account

02Save API key

After sign-up, the verify screen shows your key once. Copy it and store it somewhere safe — you can issue more later from the API Keys page.

your key

text

ki_sk_live_a8f2b39c4421•••

03Connect Claude Code

One command wires the Kireo MCP server into Claude Code.

terminal

bash

claude mcp add kireo \ --scope user \ --env KIREO_API_KEY=ki_sk_live_••• \ -- npx -y --package=@kireo/mcp-server kireo-mcp

04Connect Cursor

Add the same server to .cursor/mcp.json — both tools read and write the same memory.

.cursor/mcp.json

json

{ "mcpServers": { "kireo": { "command": "npx", "args": ["-y", "--package=@kireo/mcp-server", "kireo-mcp"], "env": { "KIREO_API_KEY": "ki_sk_live_•••" } } } }

05Write a memory

Just tell your AI to remember. It writes a typed record over MCP — no forms.

in Claude Code

text

> Remember: we use LanceDB as the vector layer ↳ memory_save { type: "decision" } ✓ Memory saved

06Recall a memory

Any connected tool recalls it — same memory, any client.

in Cursor

text

> Where did we decide to use LanceDB? ↳ memory_search { query, type: "decision" }

07Index code with Pro PRO

From any repo root, index local code into searchable type: code memories.

terminal

bash

export KIREO_API_KEY=ki_sk_live_••• npx -y -p @kireo/mcp-server kireo index ./ --repo my-app

Only symbol text + vectors leave your machine — your source stays local.

Code indexing is free during the MVP.

08Search code through MCP

Decisions and code flow through the same memory_search tool — the score your AI sees is the score the web app shows.

in Claude Code

text

> Show me the auth middleware. ↳ memory_search { query, namespace: "code-my-app" } ✓ src/auth/clerk.ts · score 0.91

09Open dashboard

Browse everything your AI remembers — memories, usage, indexed repos.

Open dashboard

Was this helpful?Next: Open dashboard