Connect your AI assistant.

EOModeller has an MCP server, the open standard AI assistants use to work with other tools. Connect Claude (or any MCP client) once, and it can read your models, answer questions about them, draw new diagrams and make changes you can see and undo. Your assistant brings its own AI; EOModeller serves and stores the model.

What you can ask

• Explain: "Using EOModeller, explain the order-system model." It reads the overview first, then the key elements and how they relate.
• Answer questions: "What depends on the payment service?" "Which classes have no description?"
• Review: "Review the billing model." It runs validation, then comments on naming, structure, coupling and gaps.
• Draw: "Create a C4 container diagram for this codebase." "Add a class diagram for the customer domain." Elements land on the canvas, laid out, ready for you to refine.
• Change: "Rename Customer to Account everywhere." "Add an Invoice class with these fields." "Apply the «entity» stereotype to every class in Domain."
• Generate code: "Generate TypeScript types from the Domain package." It maps your model's types through the languages the model declares.
• Export and import schemas: Cassandra CQL, Avro, JSON Schema, Protocol Buffers and Structurizr DSL, as files you can check in.

In Claude Code, three ready-made prompts are in the / menu: explain-model, review-model and generate-code.

Before you start

• An EOModeller account. Reading models works on every plan.
• Changing models needs a Pro plan or a trial, and the Editor role (or higher) in the model's workspace.
• An AI assistant that supports MCP: Claude Desktop, claude.ai or Claude Code, or another MCP client.

Claude Desktop and claude.ai

Nothing to copy or paste: you sign in to EOModeller from Claude.

1. In Claude, open Settings → Connectors → Add custom connector.
2. Name it EOModeller and enter the URL https://api.eomodeller.com/mcp. Leave any client ID and secret fields empty.
3. Click Connect. EOModeller opens in your browser; sign in if asked.
4. Check the page that asks for your approval. It shows the app's name and where access will be sent (for example claude.ai). Choose the access (Read only, or Read and write if you want it to make changes) and which workspaces it can use, then click Approve.
5. You're back in Claude, with the EOModeller tools available. Try "Using EOModeller, list my models."

Claude Code

Claude Code connects with a personal access token.

1. In EOModeller, open Account → API tokens, give the token a label (say, "Claude Code on my laptop") and create it. Choose write scope if Claude should change models; read is the default. Copy the eo_pat_… value: it's shown once.
2. In a terminal, add the server:

claude mcp add --transport http eomodeller https://api.eomodeller.com/mcp \
  --header "Authorization: Bearer eo_pat_..."

Then ask, for example, "Using eomodeller, explain the order-system model." Account → Connected apps shows the same command with your server's URL filled in.

Other MCP clients

Any client that supports remote MCP servers with sign-in (the MCP Authorization standard) connects like Claude Desktop: give it https://api.eomodeller.com/mcp. Clients that let you set a request header can use a token instead, as Authorization: Bearer eo_pat_…. For clients that can only start a local program, there's a small bridge, eomodeller-mcp, that the client runs and that forwards to EOModeller with your token; ask us at support@eomodeller.com and we'll point you to it.

What you'll see when it makes changes

• Changes arrive live. If you have the model open, an assistant's changes appear straight away, with a notice such as "12 changes by an AI agent (Claude Code on my laptop)".
• Revert undoes them. The notice has a Revert button that undoes the whole set at once. An assistant's changes aren't on your own Cmd+Z history; Revert is how you take them back. You can also ask the assistant to revert.
• A baseline comes first. The first time an assistant changes a model (and again after 12 hours), EOModeller saves a baseline named "Before agent changes", so there's always a known state to compare against. It's listed with the model's other baselines.
• All or nothing. Each request is checked first and applied as one step: if any part of it is invalid, nothing changes and the assistant is told why.

What it can see

Exactly what you can: the workspaces you belong to, narrowed to the workspaces you chose when you approved the app or created the token. Anything else, including models in the Trash, is "not found" to the assistant. Disconnecting the app or revoking the token cuts off access at once.

Troubleshooting

It can read my models but not change them

Changing needs three things: a Pro plan or trial, the Editor role (or higher) in the model's workspace, and write access, meaning Read and write when you approved the app or a write-scope token. To change an approval, disconnect the app under Account → Connected apps and connect again.

"Not found" for a model I can see in EOModeller

The model is in a workspace you didn't include when you approved the app or created the token. Connect again (or create a new token) with that workspace included.

The model is read-only to the assistant

Archived models, models over your plan's limits, and models in the Trash can't be changed by anyone, assistant included. Restore or unarchive the model first.

"Too many requests"

Each connection can make 60 reads and 20 sets of changes a minute. The message says how long to wait; asking for larger, fewer changes helps.

Claude Code says the token was rejected

The token has expired or been revoked, or was pasted incompletely. Create a new one under Account → API tokens and run the claude mcp add command again (remove the old entry first with claude mcp remove eomodeller).

More

• Generate code from your model: scripted exports, including the schema formats.
• C4 diagrams: the samples on that page were drawn this way.
• modelcontextprotocol.io: the MCP standard.