Give Claude a notepad you can read on your phone

Ask an AI coding agent for a plan and it ends up in a chat window or a local file. Neither is easy to read on your phone, share with a teammate or come back to next week. Pasting it somewhere else loses the formatting.

static.md fixes that with a remote MCP server. Your agent writes straight into a markdown note in your account. You open the link anywhere, edit it, and the agent reads your changes the next time it looks.

This guide covers the setup for each app, what to ask the agent once it’s connected, and what it can and can’t do.

What MCP is, in one paragraph

MCP (Model Context Protocol) is the standard way AI apps talk to outside tools. An MCP server offers a set of tools; the AI app asks you to approve them and then calls them when it needs to. static.md’s server lives at https://static.md/mcp. You sign in once with your static.md account (Google sign-in), approve the access the app asks for, and there’s no token to copy around.

Connect your AI

Claude Code

Add the server from your terminal:

claude mcp add --transport http static-md https://static.md/mcp \
  --client-id claude-code

Then run /mcp in Claude Code, pick static-md and choose Authenticate. Your browser opens static.md: sign in with Google and approve.

claude.ai and Claude Desktop

static.md is in Claude’s connectors directory. Open Settings → Connectors, choose Browse connectors and search for static.md. If your workspace doesn’t show it, choose Add custom connector instead and paste https://static.md/mcp. Select Connect, sign in on static.md and approve.

ChatGPT

ChatGPT takes custom MCP connectors in developer mode:

1. Open Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
2. Back in Apps & Connectors, choose Create, name it static.md, set authentication to OAuth and paste https://static.md/mcp.
3. Sign in on static.md and approve, then switch static.md on from the + menu in a chat.

ChatGPT renames these menus often; if a label differs, look for “developer mode” and “custom connector”.

Cursor and VS Code

Both have one-click installs on the agents page. By hand, add this to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "static-md": { "url": "https://static.md/mcp" }
  }
}

or this to .vscode/mcp.json:

{
  "servers": {
    "static-md": { "type": "http", "url": "https://static.md/mcp" }
  }
}

The editor asks you to sign in to static-md the first time it uses it.

Scripts and other clients

Clients without sign-in can use a personal access token as a bearer token. Create one under Connected apps & tokens and treat it like a password.

What to ask once it’s connected

The agent gets tools to list, read, create and update notes; list, read and create whiteboards; and change who can open or edit something. You don’t call them yourself. Just ask in plain words:

• “Write the plan for this refactor into a new static.md note and give me the link.”
• “Read my note Weekly priorities and tell me what’s overdue.”
• “Add today’s decisions to the end of the meeting note.”
• “Sketch the request flow as a whiteboard.”

Edits merge live with whoever else has the note open, so you can watch the agent type into a note on your phone while it works on your laptop.

A few habits that work well:

• Keep one running note per project. Ask the agent to append a dated section each session. It becomes a log both of you can read.
• Write requests back into the note. Next session, say “read the project note and do what I asked at the bottom”.
• Share the link, not the chat. A note link opens in any browser, with Mermaid diagrams and code highlighting rendered.

What the agent can and can’t do

You stay in control of the access:

• You approve it. Every app signs in through static.md and you approve it with your Google account. Nothing connects on its own.
• Access is scoped. Reading and writing are separate, and so are notes and whiteboards. An app gets only what you approved.
• Sharing is its own permission. Changing who can open a note needs separate approval, and agents are told to use it only when you ask directly.
• It follows your sharing settings. An agent opens exactly what you could open in your browser.
• Encrypted notes stay private. static.md can’t read notes you encrypt end-to-end, and neither can an agent, unless you give it the key through the open-source E2E bridge on your computer (Pro).
• Notes are data, not instructions. Note text reaches the agent fenced as user content, so instructions hidden in a note don’t read as commands.
• You can revoke it. Disconnect an app or delete a token at any time; it stops working within seconds.

One thing to know: a note the agent creates opens to anyone with the link, like any new static.md note. Restrict it in the note’s sharing settings (Pro) if it’s private, or ask the agent to.

Limits

The free plan includes 50 tool calls a day and 10 new notes or whiteboards an hour. Pro raises that to 5,000 calls a day with no limit on new notes, and adds end-to-end encryption and access control. Details on pricing.

Try it

Connect your AI from static.md/agents, or try the editor first at static.md/md/demo, no account needed.