Custom add-on

What is a custom MCP tool?

A custom MCP tool is an add-on for MCP Self-hosted that we build on request, for a system our standard add-ons do not cover. Think of a niche CRM, your ERP or an internal API. Once it is built, it runs on your MCP Self-hosted license like every other add-on: Claude sees it as a set of tools, and you decide who gets access and whether it can write. This product is the request form for such a tool. It requires MCP Self-hosted.

Do you need a custom tool, or does an add-on exist?

Check the existing add-ons first. They cover the systems most teams connect Claude to:

• WooCommerce: orders, products and customers, with several stores on one add-on.
• WordPress: posts, pages, media, menus, users and settings, with several sites on one add-on.
• Database: MySQL/MariaDB-compatible or PostgreSQL, with list tables, describe table and run query.
• Missive: conversations, messages, contacts and tasks in a shared inbox.
• Brain: a markdown knowledge base with optional git versioning.
• Lightspeed eCom: read-only access to products, orders and customers in the C-Series API.

If your system sits behind one of these, a standard add-on is faster to get running. If it does not, a custom MCP tool is the route.

How does the custom MCP tool process work?

The process has three steps:

1. You describe the system and what your AI client should be able to do with it.
2. We reply with a proposal covering scope, read and write access, price and delivery time.
3. Once agreed, we build the add-on and add it to your MCP Self-hosted license, just like any other add-on.

Nothing is built before you agree to the proposal. The price is part of that proposal and depends on the scope you describe.

What should you include in your request?

The proposal can only be as precise as the description. Two things make the biggest difference.

First, the system itself: what it is, and how it can be reached from outside. Second, the tasks. Write them as the questions or actions you want Claude to handle, for example “look up a customer’s open invoices” or “list orders that have not shipped”. Separate the tasks that only read data from the ones that change it. That split decides the read and write scope in the proposal.

What happens once the custom MCP tool is on your license?

From that moment it behaves like every other add-on. MCP Self-hosted checks its license with AddoCore on start and periodically, and receives entitlements: the list of tools that are allowed to load. With the new entitlement in place, the tool appears. Without it, the tool is not exposed at all.

There is no reinstall and no new download. Refresh the license by hand, or wait for the next check:

docker exec mcp-core npm run refresh-license

Is a custom MCP tool read-only?

Yes, by default. A custom add-on follows the same rules as every MCP Self-hosted tool: it starts read-only, and write access is something you switch on yourself. You set that per tool and per seat, so one person can get write access while the rest of the team only reads.

Access per person works through seats. Each seat has its own token and its own subset of tools, so you choose which teammates see the custom tool at all. MCP Self-hosted includes 3 seats; an extra seat adds one more.

Who is a custom tool for?

It suits teams that already use MCP Self-hosted for their standard systems and have one system left that Claude cannot reach yet. Running everything through one self-hosted server keeps one license, one set of seats and one update command, ./mcp-update.sh, for all tools.

Need a custom MCP tool for your own system? Describe it in the request form and we reply with a proposal for scope, access, price and delivery time.