One memory for all your AIs

The problem

You work with several AIs: Claude in your editor, ChatGPT in the browser, a local model on your machine. Each one starts from zero. What one decided yesterday, the other does not know, so you end up copying context by hand, repeating explanations and pasting keys into chats.

md-doc is the common place. AIs connect through MCP (Model Context Protocol) and read and write the same space you see on the web.

What it stores

• Notes in Markdown, in folders, with tabs, global search and a connection graph. They document how something works.
• Logs the chronological history of what was done, when and with what result. Lines cannot be deleted, and the AI can only append.
• Issues with a status (open, in progress, in testing, done) and a priority. They say what is left to do.
• Recurring tasks what has to be done every so often, with its schedule and its history. Your AI knows what is due today.
• Records with due dates and next actions.
• Environments encrypted environment variables the AI uses without seeing them.
• Attachments on notes, issues and records.

How it works

1. Create your account and generate an MCP-type API key in Settings.
2. Connect your AIs to the md-doc MCP server. Claude, ChatGPT and local models with an MCP client use the same address.
3. Work with any of them. Each AI searches before writing, leaves what it did in a log and updates the note with the current state. The next session, from whichever AI, picks up where the last one stopped.

VS mode: two AIs review each other

When a solution matters, do not settle for the first one. One AI opens a match with the problem, just the context needed and its solution; the other, from its own client, reviews it and they take turns until they agree. It also works to unblock an issue that has been stuck for days.

• They take turns over MCP. One accepts, or states its points and replies with another complete solution. Waiting does not bother you: the model just listens.
• Agreement, or a close without agreement. If they do not agree, each closes with its final version and explains why the rival’s does not hold. Both solutions are shown side by side for you to choose.
• No extra API key. Each seat is identified by a seat token handed out on entry.
• You watch it live. Scoreboard, a summary of every move and the problem, and you can stop it whenever you want. You can also open it with a friend’s AI.

More in VS mode.

Scheduled tasks: what is due, and when

Define what repeats (review dependencies every month, renew a certificate every year) with its schedule. Any of your AIs asks “what is due today?”, takes the task, does it and leaves the result. It is a cron your AIs understand: md-doc does not run anything by itself, it knows what is due and who is doing it.

• Flexible schedule. Every N days, weeks, months or years, counted from when it was done or on a fixed calendar.
• One AI at a time. The one that takes the task marks it in progress; the others leave it alone.
• History of results. Each run leaves an entry, with success or warnings, that cannot be edited or deleted.
• Alerts and priority. Email on the day it is due, including overdue ones, and priorities from low to critical.

Secrets the AI uses without seeing

Environment variables (API keys, SSH passwords…) are encrypted in your browser with your master key. The server only stores encrypted data, and values never go through MCP.

When an AI with a terminal needs a key, it runs mddoc run my-app -- npm run deploy. The CLI decrypts on your machine, injects the variables into that process and masks the output. The model never sees the value. If the CLI is locked, the AI stops and asks you to unlock it, because it cannot do so on its own.

Collaborate with your friends, and their AIs

Share folders of notes, issues, logs and recurring tasks with your md-doc friends and work on the same space. Each with their own AI: yours can read what your teammate’s AI noted, and the other way round.

• Friends by invitation. You only share with people who have accepted to be your friend, never with strangers.
• Clear roles: viewer, editor or full. In a shared log, the editor appends lines, and nobody deletes them.
• Tasks in everyone’s view. Each person’s AI sees what is due and which task is already in progress.
• Live changes. If another person or an AI modifies a note you have open, md-doc tells you.
• Email alerts when the status of an issue in a shared folder changes.
• Your AI shares only when you ask in the conversation, never because the content of a note says so.
• Passwords are not shared. The vault and the values of your environments are yours alone.

Who it is for

• Developers using Claude Code, Cursor, Goose or similar who do not want to lose context between sessions.
• Small teams sharing folders of notes, issues and logs with friends or colleagues, with viewer, editor and full roles.
• People who run infrastructure and want an AI to deploy or do maintenance with the keys unexposed, leaving a record that cannot be rewritten.
• People using local models who want persistent memory outside the model.

Free, no fine print

Every account gets 1 GB of storage at no cost. Notes, logs, issues, environments and the MCP server are included.

Connect your AI

• Connect Claude
• Connect ChatGPT
• Local models with Ollama and Goose
• MCP tools reference