Your AI agent drawsand maintains your diagrams.
The agent in your editor already knows the architecture. Connected to Datadef over MCP, it creates the diagram, edits it as the system changes, and exports it when you need a file, all from the chat you already have open.
Part of Pro and Team · 7-day free trial · No credit card
Claude Code
Cursor
Copilot
Codex
Gemini CLIHow it works
One key, one command, then just ask
Datadef runs a remote MCP server. There is nothing to host and nothing to update.
Create an API key
On the MCP & integrations page in settings. The key is shown once, and the setup snippet comes prefilled with it. API access is part of Pro and Team, and the trial covers it.
Connect your client
One command for Claude Code, one-click installs for Cursor and VS Code, snippets for Claude Desktop, Codex, and Gemini CLI. ChatGPT connects over OAuth, no key pasted.
Ask in plain language
Describe the system, or let the agent describe the repo it is sitting in. The diagram lands in your workspace, editable, shareable, and ready to embed.
Not a chat artifact.
Agent-built diagrams are real canvases in your workspace, drawn to the same design standard as the in-app generator. Nothing lives only in the conversation.
The endpoint is https://datadef.io/mcp. Open it in a browser and it serves the connect instructions.
Setup
One page in settings
Create the key, pick your client, copy the prefilled command. This is the actual page.
In practice
The tools are the workflow
Draw it, change it, ship the image, keep it honest against the repo. Each step is a tool the agent calls for you.
create_diagram turns a description into a finished, editable diagram and returns a preview
edit_diagram applies plain-language changes: add nodes, rewire connections, relabel layers
export_diagram returns the image inline plus a download link that needs no API key
repo_status · repo_refresh check and re-sync diagrams linked to a repository
> Draw the checkout flow: API, orders DB, Kafka, payments worker.
create_diagram
Your diagram is ready. Open and edit: datadef.io/project/…
> We put a retry queue between Kafka and the worker. Add it.
edit_diagram
Retry Queue added between Kafka and the payments worker · updated preview attached
Honest answers
Which clients work?▼
Claude Code, Cursor, VS Code, Claude Desktop, Codex, Gemini CLI, and ChatGPT. Anything else that speaks MCP over Streamable HTTP connects the same way; the settings page keeps a prefilled setup snippet per client.
Where does the API key come from?▼
From the MCP & integrations page in settings. Create a key and copy it once, it is shown a single time, then paste it into the setup command for your client. ChatGPT is the exception: it connects with an OAuth sign-in and no pasted key.
Is MCP on the free plan?▼
API access is part of Pro and Team. Every new account starts with a 7-day trial of the full product, no card required, so you can connect an agent on day one.
What can the agent actually do?▼
Create diagrams from a description, build or edit them node by node, change them with plain-language instructions, list and read what is in your workspace, export PNG or JPEG images, and check or refresh repo-linked diagrams with repo_status and repo_refresh. Agent-built diagrams follow the same design standard as the in-app generator.
Does it watch my repository on its own?▼
No. Over MCP, tools run when you ask, and nothing moves between sessions. Diagrams that regenerate daily without an agent in the loop are repository sync, a separate feature that pairs with these tools.
How repo sync worksDo the diagrams stay in the chat?▼
No. Every diagram the agent creates is a project in your Datadef workspace: open it on the canvas, hand-edit it, share it, embed it. The chat is just where the work got asked for.
Give your agent a canvas
Connect it once and the diagrams get drawn, updated, and exported from the chat you already work in. Every new account starts with a 7-day trial of everything, no card needed.