MCP server

MCP (Model Context Protocol) is the standard way an AI assistant talks to an outside service. Once Dodil is connected, you can just ask your assistant to do things — “list my buckets”, “show my apps’ logs” — and it calls Dodil for you. No code, no API keys.

There are two ways to connect, and picking the right one takes about ten seconds.

Which one do I want?

	Remote server — `mcp.dodil.io`	Local server — `dodil mcp`
Who it’s for	Anyone using Claude in a browser or the Claude desktop app, ChatGPT, Perplexity, Gemini — anything that runs “in the cloud”	Developers using a coding agent on their own machine (Claude Code, Cursor, VS Code Copilot, Zed…)
Do I need to install anything?	No. You paste one web address into a settings screen	Yes — the `dodil` CLI (one command)
How do I log in?	A browser window opens, you log in to Dodil, done	You’re already logged in from `dodil auth login`
Where does it run?	On Dodil’s servers	On your own computer
Set-up time	~1 minute	~2 minutes, once, for all your agents

Rule of thumb: if your assistant lives in a browser or a phone app, use the remote server. If it lives in your terminal or code editor, use the local server.

Jargon, once: a server here isn’t a machine you have to run — it’s just the thing your assistant connects to. The remote one is already running; the local one is your dodil command wearing a different hat.

Get connected

Pick your assistant:

Claude (web & desktop app) — with screenshots of the exact dialog
Claude Code — the terminal agent; use the local server here
Codex Desktop & CLI — use the local server here
Cursor, VS Code, Perplexity, Gemini and others
ChatGPT (limited) — hosted, read/fetch only on Pro; not recommended for builds or deployment

Something not working? → Troubleshooting

What you get, either way

Every Dodil product, as tools. Ignite (apps), DataK³ (buckets, tables, vector search), Git, Registry and Scriptum are all exposed as named tools — ignite_app_list, k3_bucket_get, git_repo_list, registry_repo_list, scriptum_thread_run, and so on. Tools mirror the CLI one-for-one: flags become named parameters, positional arguments go in an args array, and output is JSON.
Your own permissions. The assistant acts as you. It can only see and change what your Dodil account is already allowed to see and change — see Roles and Policies.
Plain-English use. You never type a tool name yourself. You describe what you want; the assistant chooses the tool.

Try it once you’re connected

Type these straight into the chat:

“List my Dodil buckets.”
“Show me my Ignite apps and which ones are running.”
“Create a new Dodil git repository called invoices-api.”
“What container images are in my Dodil registry?”
“Show the last 50 log lines for my Ignite app checkout.”

The assistant will ask before doing anything that changes or deletes something — approve or reject it in the chat.