memory-wiki-mcp
v1.5.4
Published
MCP server for memory.wiki: tools for AI agents to create, edit, organize, search, share, version, and paste-anywhere Markdown documents.
Maintainers
Readme
memory-wiki-mcp
MCP server for memory.wiki. Lets any AI tool create, read, update, search, and paste-anywhere your Markdown documents with permanent shareable URLs.
Works with Claude Code, Claude Desktop, Cursor, and any Model Context Protocol compatible client.
Two ways to connect:
- Hosted HTTP MCP (recommended for Claude Web, Cursor, etc.). No install. Just add
https://memory.wiki/api/mcpin your client's MCP / Connectors settings. Exposes the full tool set.- Local stdio MCP (this npm package). For Claude Desktop and Claude Code. See setup below.
Quick Start
1. Login (one-time)
npx memory-wiki-cli loginOpens your browser for OAuth. Credentials are stored locally in ~/.memory.wiki/.
2. Add to your AI tool
Claude Code / Cursor — create .mcp.json in your project root:
{
"mcpServers": {
"memory.wiki": {
"command": "npx",
"args": ["memory-wiki-mcp"]
}
}
}Claude Desktop — add to claude_desktop_config.json:
{
"mcpServers": {
"memory.wiki": {
"command": "npx",
"args": ["memory-wiki-mcp"]
}
}
}No API keys or environment variables needed. Authentication is handled via mw login (run once before starting the MCP).
Tools
| Tool | Description | Auth |
|------|-------------|------|
| mw_create | Create a new document and get a shareable URL | Optional |
| mw_read | Fetch document content by ID or URL | No |
| mw_update | Update an existing document's content | Edit token (auto-managed) |
| mw_list | List all your documents with metadata | Yes |
| mw_publish | Toggle a document between public and private | Edit token (auto-managed) |
| mw_delete | Soft-delete or permanently delete a document | Edit token (auto-managed) |
| mw_for_ai | Return the canonical "Use https://memory.wiki/<id> as my context." sentence for any id / URL / @username / bundle. Lets one AI hand a doc to another. | No |
(Hosted HTTP MCP at memory.wiki/api/mcp exposes the full set: search, version history, folders, bundles, hub queries, AI re-analyse triggers, etc.)
What You Can Do
You: "Create a document with my meeting notes"
AI: mw_create → https://memory.wiki/abc123 (URL copied!)
You: "List my documents"
AI: mw_list → 8 documents found
You: "Read the system design doc"
AI: mw_read → (full markdown content)
You: "Update it with the new architecture section"
AI: mw_update → Document updated
You: "Make it private"
AI: mw_publish (published: false) → Now private
You: "Delete the draft"
AI: mw_delete → Moved to trashCross-AI Workflow
memory.wiki URLs work as context across AI conversations. Each URL is a single canonical link that fetches as clean markdown for any AI.
You (in Claude): "Summarize the research at memory.wiki/abc123"
AI: mw_read fetches the document, summarises
You (in Claude Code): "Get me the for-AI paste for that doc so I can give it to Cursor"
AI: mw_for_ai abc123 returns: Use https://memory.wiki/abc123 as my context.
You (paste that line into Cursor): Cursor now has the same context.
You (in Claude): "Update memory.wiki/abc123 with the improvements"
AI: mw_update writes back to the same URL.URL forms
| Form | Pattern | When |
| --- | --- | --- |
| Doc | memory.wiki/<id> | Single document, 12-char nanoid |
| Bundle | memory.wiki/b/<id> | Curated set of docs |
| Hub | memory.wiki/@<username> | Full user hub (all public docs + bundles) |
| Raw | memory.wiki/<id>.md | Explicit markdown payload |
The mw_for_ai tool wraps any of these into the canonical "Use ... as my context." paste sentence so AI clients hand them off to each other without inventing their own phrasing.
How Authentication Works
The MCP server shares credentials with the mw CLI:
mw loginopens your browser for Google / GitHub OAuth- JWT token is stored locally in
~/.memory.wiki/config.json - Edit tokens for each document are stored in
~/.memory.wiki/tokens.json - All API requests use
Authorization: Bearerheaders - Tokens auto-refresh when expired (clear error message if re-login needed)
No email spoofing possible. All requests are authenticated via JWT.
Features
- Permanent URLs. Every document gets a short URL (
memory.wiki/<id>) that never expires. - Auto-managed edit tokens. Create a doc, get edit access automatically.
- Public or private. Toggle visibility with
mw_publish. - Markdown rendering. Documents render with syntax highlighting, KaTeX math, Mermaid diagrams.
- Version history. All edits are tracked.
- Zero config. Just
npx memory-wiki-mcp, no API keys needed.
Other Channels
memory.wiki is available everywhere:
| Channel | Install |
|---------|---------|
| Web Editor | Just open the URL |
| CLI | npm install -g memory-wiki-cli |
| VS Code Extension | Download .vsix from Plugins page |
| Chrome Extension | Download from Plugins page |
| Mac Desktop App | Download .dmg from Plugins page |
Links
- Website: memory.wiki
- Plugins: memory.wiki/plugins
- API Docs: memory.wiki/docs
- GitHub: github.com/raymindai/memory-wiki
License
MIT
