Skip to main content
This guide gets you a working OneLamp setup: an account, your personal MCP endpoint, and a tool that loads and saves your context automatically.

1. Create your account

Go to app.onelamp.ai and sign in with GitHub, Google, or a one-time email code. This is the only account you’ll need — OneLamp is its own identity provider, so your AI clients authorize against it directly.

2. Copy your MCP endpoint

After signing in, the Connect home shows your personal MCP endpoint:
https://api.onelamp.ai/mcp
The endpoint URL is the same for everyone, but it’s personal and authenticated — every request is scoped to the signed-in user. There are no API keys to copy or rotate; clients sign in with OneLamp via OAuth.

3. Connect a client

It’s one endpoint, so connecting is the same everywhere: add https://api.onelamp.ai/mcp to your client as a remote MCP server. The first time it connects, it opens OneLamp in your browser to sign in and authorize — no API keys. After that, the token is cached and refreshed automatically. A few clients add more on top (Claude Code, Codex, and Cursor ship lifecycle hooks that auto-load and auto-save context). The Connect your tool page has the copy-paste setup for each.

Connect your tool

Universal setup plus the client-specific bits for Claude Code, Codex, Cursor, VS Code, and ChatGPT — each has its own tab.

4. Seed your context (one-time)

Connecting captures what your tools learn going forward. To also pull in what a tool already knows about you, run a one-time seed:
1

Claude Code

Run /ol (alias of /onelamp). It scans this repo’s CLAUDE.md, .cursorrules, or AGENTS.md and saves them with save_context.
2

Any other tool

Tell it: “Fetch https://app.onelamp.ai/setup.md and follow it.” It seeds your context from what the tool already knows and adds OneLamp to the tool’s system prompt so it keeps loading and saving context — which is how hookless browser tools (Gemini, Perplexity, ChatGPT web) stay in sync.

5. Verify the connection

Once connected, your tool has the context tools available. Ask it to run a quick check:
1

Save something durable

Tell your tool: “Remember that I deploy on Fridays.” It should call save_context and confirm with an id.
2

Read it back

In a fresh session, ask: “When do I deploy?” The tool calls get_context and retrieves the fact you saved — even from a different tool.
3

Export your data (anytime)

Your data is always yours to take. From your Account in the OneLamp app, export your full store as portable JSON on demand — no queue, no lock-in.
If the tool saved a fact in one tool and retrieved it in another, OneLamp is working — every tool you connect now uses what the others have learned, with no re-explaining.

What’s next

Understand context memory

How saving, deduping, and ranked retrieval work under the hood.

Build your Library

Ingest documents and URLs into an interlinked Library your tools can query.

Connect your tools

Pull in documents from Notion, Drive, Slack, and other MCP servers.

Browse the tools

The full reference for every tool your AI tool can call.