# Connect your agent

> Point an MCP client at OceanDB and authorize it once.

OceanDB is reached over MCP — the Model Context Protocol. Any MCP-capable agent
can connect; there's no SDK to install and no keys to paste.

## The endpoint

```
https://app.oceandb.ai/api/mcp
```

Add this as an MCP server in your agent. That's the whole configuration —
and it's the same memory from every client, because the memory belongs to
you, not to any one agent.

## Per client

| Client | How |
| --- | --- |
| Claude Code | `claude mcp add --transport http oceandb https://app.oceandb.ai/api/mcp` |
| Claude Desktop | Settings → Connectors → add the endpoint URL |
| Codex CLI | Add the endpoint as an MCP server in `~/.codex/config.toml` |
| Cursor | Add the endpoint to your MCP configuration |
| ChatGPT (work) | Settings → Connectors → add a custom MCP server |
| Any MCP client | Point it at the endpoint; OAuth does the rest |

One endpoint, one mechanism — every row is the same connection.
For current, client-specific setup and limitations, use the
[integration guides](/integrations).

Simpler still: hand your agent the one-liner and let it install itself —

```
Install my memory: https://oceandb.ai/docs/connect
```

## Authorize once

OceanDB is its own OAuth 2.1 authorization server, with dynamic client
registration. You don't manage credentials by hand — the flow runs itself:

1. Your client connects and is asked to authorize.
2. A browser opens to OceanDB. Sign in and approve access.
3. Your agent receives a token and can now reach your memory.

Behind the scenes, the unauthorized request returns the address of the
authorization server, your client registers itself, and the standard OAuth
handshake completes. You just click approve.

New account? Sign-in is a code sent to your email — approve once and the
connection is yours.

## What your agent gets

Three tools, all under a least-privilege role scoped to you:

- **`remember`** — structured capture: one call saves an entry, applies tags,
  and embeds it for semantic recall.
- **`recall`** — hybrid retrieval: semantic similarity fused with keyword
  search across your memory.
- **`execute_sql`** — the full language underneath: a single SQL statement
  for everything the shortcuts don't cover. See
  [The execute_sql contract](/docs/execute-sql).

A fourth, `onboarding_status`, reports where the memory stands and what to
set up next — a freshly connected agent calls it and follows the steps.

It also gets three reference resources it can read on demand:

- `oceandb://docs/ocean` — the full capture-and-retrieve manual.
- `oceandb://docs/dreamer` — the nightly Dreamer playbook.
- `oceandb://schema` — the live schema.

## Signing in as a human

To review what your agent remembers, open the dashboard at
[app.oceandb.ai](https://app.oceandb.ai) and sign in with a magic link. No
password to keep. See [Dashboard overview](/docs/dashboard).

## After connecting

Have your agent call `onboarding_status` and follow its `next_step` — it
walks the rest of this path: the master prompt, a first memory, the Dreamer.

Next: install [the master prompt](/docs/master-prompt), then capture
[your first memory](/docs/getting-started).