Connect your AI agent to Aleta

The Aleta MCP server lets an AI agent read and work with your Aleta data directly. Setup takes about a minute.

You need:

An Aleta login.
An agent app that supports remote MCP servers — Claude, ChatGPT, Cursor, VS Code and most others do.
The server URL: https://mcp.aleta.io/mcp

The screenshots below are from Claude. Other apps use different wording — look for MCP, Connectors, Tools or Integrations in settings — but the steps are the same.

1. Add the server

In Claude, as of September 2026, this lives under Customize → Connectors. Click Add, then paste https://mcp.aleta.io/mcp as the URL. The name is up to you — we suggest Aleta. Continue to the next screen.

The Add custom connector dialog, with the name Aleta and the URL https://mcp.aleta.io/mcp — The name and the URL are the only two values you enter.

2. Accept the detected settings

The next screen decides how the agent signs in, and it fills itself in: authentication comes up as Always required and the OAuth client as Anthropic's hosted client metadata, both marked Detected. Leave them alone and add the connector.

Authentication set to Always required and OAuth client set to Anthropic's hosted client metadata, both marked Detected — Nothing here needs changing — the defaults are already correct for Aleta.

3. Sign in

The connector now exists but is not connected yet. Click Connect and sign in with your Aleta login. The agent gets exactly the access your own Aleta user has — nothing more.

The Aleta connector showing 'You are not connected to Aleta yet' above a Connect button — Until you sign in, the connector is listed but inactive.

4. Check the tool permissions

The connector lists its tools in two groups. The read-only tools — reading your data, and looking up the API spec — are set to Always allow. The write/delete tools, which create, change or remove data, are set to Needs approval: the agent has to ask you each time before it uses one.

Tool permissions: read-only tools set to Always allow, write/delete tools set to Needs approval — The default setup. Nothing is created or changed without you confirming it first.

That's it — the connection works. Ask your agent to use the Aleta MCP and dive into your data.

Read-only setup

If you would rather the agent could never change anything, set the write/delete group to Blocked. The read-only tools stay on, so the agent can still answer questions about your data — it simply has no way to modify it.

Tool permissions with the write/delete tools group set to Blocked and those tools greyed out — Read-only. With the write/delete tools blocked, the agent cannot create or change data even if it is asked to.

We recommend this for anyone who uses Aleta mainly to look things up, and for organisations that want to rule out accidental changes made through an agent. Approving a write tool once, in a long conversation, is easy to do without meaning to; blocking removes the risk entirely.

These settings live in your own agent app, so each person can choose their own. If you administer Claude for your organisation, you can set the limits for everyone centrally instead — see below.

Setting the limits for your whole organisation

Everything above is a personal setting, which means it can be changed back just as easily. An administrator can set the same limits centrally, once, for every member of the organisation.

In Claude, as of September 2026, this lives under Organization Settings → Connectors. Click the Aleta MCP connector to open Connector details, select the Configuration tab, and scroll to the bottom. The tools are listed in the same two groups your members see.

What you choose here is a ceiling rather than a fixed value: members can make their own setting stricter, never looser. Setting the write/delete tools to Ask leaves each one reading Users can choose: Ask or Block — the agent must get approval before it writes anything, and no one can turn that off.

The Configuration tab of Connector details, with the three write/delete tools set to Ask and each noting that users can choose Ask or Block — Approval enforced. Members can still tighten a tool to Blocked, but Always allow is no longer available to them.

Choosing Blocked instead makes the read-only setup organisation-wide: each write/delete tool greys out and reads Tool blocked for all users. Agents across the organisation can then read Aleta data and nothing else.

The same tab with the three write/delete tools set to Blocked, each greyed out and marked Tool blocked for all users — Read-only for everyone. No member setting can re-enable a blocked tool.

One row deserves a second look: Default restriction for unconfigured tools covers any tool without a setting of its own, including tools the connector gains later. The read-only tools sit under it — they show Uses default — so if you leave it at Always allow, a tool we add in future arrives unrestricted. Set it to Ask if you would rather approve new tools before anyone uses them.

These restrictions apply to Claude.ai, Claude Desktop and Claude Code on the web. They do not cover Claude Code CLI — even when it connects to the same connector — Desktop Extensions or the Claude API, each of which is configured elsewhere. Blocking the write tools here does not, on its own, make your organisation read-only everywhere.