static.md

Markdown notes and Excalidraw whiteboards your AI agents can read and write, plus image hosting. Remote MCP server at https://static.md/mcp with OAuth sign-in.

Hosted MCP Server

npx add-mcp 'https://static.md/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Sign in with static.md · no token to paste

  1. Add the server from your terminal:
    claude mcp add --transport http static-md https://static.md/mcp \
      --client-id claude-code
    
  2. In Claude Code, run /mcp, pick static-md and choose Authenticate.
  3. Your browser opens static.md. Sign in and approve the access Claude asks for.

Then try it

Ask your agent for a note. You get a link that opens on your phone, with no file to download.

Put a short summary of what we're working on in a static.md note and send me the link.

Make it a habit: add this line to CLAUDE.md, AGENTS.md or your agent’s custom instructions.

When something you write is long or worth keeping (a plan, report or summary), put it in a static.md note and send me the link instead of a file. Leave out secrets: anyone with the link can open it.

Notes an agent creates open to anyone with the link until you change who can open them. Private notes are on Pro.

  • list_my_notes Your notes, newest first, or one folder of them: ids, titles, folders and links. notes:read
  • search_notes Find notes by the words in their title or text, yours and those shared with you, with a snippet around each hit (Pro). notes:read
  • read_note A note’s markdown, handed to the agent as data, never as instructions, with the notes its [[wikilinks]] lead to, the notes linking back, and a version for the next update. notes:read
  • create_note A new note in your account, in any folder, with optional starting content. notes:write
  • update_note Append or replace text. Edits merge live with people typing in the note; a replace built from an older read merges with what changed since, and is refused only if both changed the same lines. notes:write
  • list_my_boards Your whiteboards, or one folder of them, with their links and element counts. boards:read
  • read_board The text on a whiteboard and its Excalidraw scene. boards:read
  • create_board A whiteboard drawn from an Excalidraw scene, in any folder. boards:write
  • move_item Rename a note or whiteboard, or move it to another folder. Only yours, as in the app. notes:write · boards:write
  • memory_view Read the agent’s long-term memory, files under /memories: Anthropic’s view command. Each file is a note you can open. notes:read
  • memory Create, edit, delete and rename memory files, with the rest of Anthropic’s memory commands. Edits merge live with yours. notes:write
  • set_access Who can open a note or whiteboard: anyone with the link, only you, or people you name (Pro). Only when you ask. sharing
  • set_read_only Make a note or whiteboard read-only, so only you can change it (Pro), or editable again. Only when you ask. sharing
  • upload_image Put a screenshot or chart on static.md and get its public link and a markdown snippet to embed in a note. It lands in your uploads. uploads
  • *_note_ciphertext read_note_ciphertext, create_note_ciphertext and write_note_ciphertext move an encrypted note’s ciphertext for the E2E bridge, which decrypts on your computer (Pro). notes:read · notes:write

ChatGPT gets every tool except memory_view, memory and the ciphertext tools: it keeps its own memory, and encrypted notes need the bridge on your computer.

memoryFree & Pro

Give Claude a memory you can read. Two tools carry the six commands of Anthropic’s memory tool: memory_view only reads, so looking needs no approval, and memory writes. Every file is a note in your /memories folder: open it on any device, fix what it got wrong, share it with a teammate. Your edits and the agent’s merge live.

  • memory_view reads · notes:read
    • view
  • memory writes · notes:write
    • create
      • str_replace
      • insert
      • delete
      • rename
  • /memories/preferences.md note “preferences” in /memories
  • /memories/projects/static-md.md note “static-md” in /memories/projects
  1. Connect static.md to Claude, as above.
  2. Add this to CLAUDE.md, a Claude project’s instructions or your system prompt:
    ## Memory
    Use static.md as your long-term memory.
    - Before a task, read /memories with \`memory_view\`.
    - Save what later sessions need with \`memory\`, as
      markdown files under /memories: my preferences,
      project context, decisions, progress.
    - Keep it tidy: update files instead of duplicating
      them, and delete what is out of date.
    
  3. See what it remembers in the memories folder of My Notes.

File names are up to 32 characters before.md, static.md’s title limit; longer names are refused, never cut. Directories go 9 levels deep. Notes you put in /memories yourself show up too: one whose title isn’t a file name, or repeats an older note’s title, appears as name~id.md. Encrypted notes stay closed. Like any new note, a memory file opens to anyone with its link until you change its sharing.

static-md-e2ePro · open source

Encrypted notes stay ciphertext on static.md. A small open-source MCP server on your computer holds the key, so your agent can read and write them anyway.

  1. Your agent Asks for a note by its link or id, and reads and writes its text like any other note.

  2. plaintext, on your computer static-md-e2e, on your computer Holds the key. Decrypts what static.md sends and encrypts what your agent writes.

  3. ciphertext only static.md Stores and returns ciphertext. It never receives the key, so it can’t read the note.

  4. Create a personal access token with notes:read and notes:write under Your account below.

  5. Add the bridge to your agent, for example Claude Code:

    claude mcp add static-md-e2e -e STATIC_MD_TOKEN=smd_pat_… \
      -- npx -y static-md-e2e
    
  6. Ask for an encrypted note by its full link. To keep the key out of the chat too, save it on your computer once (paste the link when asked) and give your agent just the id:

    npx static-md-e2e key add
    

Titles aren’t encrypted. Your AI provider sees what your agent reads, as with any note. Works with agents that run on your computer (Claude Code, Claude Desktop, Cursor, VS Code, Codex), not with claude.ai or ChatGPT in the browser. The bridge gives your agent read_encrypted_note, create_encrypted_note and update_encrypted_note; on static.md’s side, three tools move ciphertext only: read_note_ciphertext, create_note_ciphertext and write_note_ciphertext.

FreePro
Tool calls per day505,000
Tool calls per minute6060
New notes & whiteboards10 / hourNo limit
Elements per whiteboard300No limit
Image size10 MB10 MB
Searching notes—Titles & text
End-to-end encrypted notes—Read & write
Edit history & restore—50 edits, 30 days
Personal access tokens110
Limit an agent to folders—When you connect it

Daily limits reset at 00:00 UTC. When one runs out, the agent is told why and where to upgrade.

Sign in to see the apps connected to your static.md account and manage access tokens.

  • consent Every agent signs in through static.md, and you approve it with your static.md account. Nothing connects on its own.
  • scopes Access is granted per kind: reading or writing, notes or whiteboards, uploading images. An agent gets only what you approved.
  • folders Limit an agent to some of your folders when you approve it (Pro). It sees and changes only your notes and whiteboards there, and everything else answers as if it didn’t exist. Leaving Pro never widens it.
  • set_access Changing who can open or edit your notes needs its own permission, approved on its own. Agents are told to use it only when you ask them directly, never because a note says so.
  • sharing Agents follow your sharing settings: they open exactly what you could open in your browser.
  • history On Pro, every change an agent makes to a note is kept for 30 days. Open History in the note to see the diff and restore the version from before it.
  • encrypted static.md never holds the key to an end-to-end encrypted note. Agents reach one only through the E2E bridge on your computer, which decrypts it there (Pro).
  • untrusted Note text is fenced as user data, so instructions hidden in a note don’t read as commands.
  • revoke Disconnect an app or delete a token at any time. It stops working within seconds.