mmm-mcp-server
# Matryoshka Mind Map MCP Server
English | [日本語](https://github.com/nhorie-canal/mmm_mcp/blob/main/README.ja.md)
**Work with your AI in maps, not text.** Your usual AI, such as Claude, writes nested to-dos and Mermaid diagrams into [Matryoshka Mind Map](https://matryoshkamap.web.app). You look, edit and check things off in the app, and the AI reads your changes back.

**The app (free to start)**: [Web](https://matryoshkamap.web.app) · [App Store](https://apps.apple.com/app/id1613648306) · [Google Play](https://play.google.com/store/apps/details?id=jp.spheres.mm)
## How it works
```mermaid
sequenceDiagram
actor You
participant M as Your mmm map
participant AI as Your AI
Note over You,AI: Instead of long text back and forth, you work through the map
You-)AI: A short request: "Turn the meeting into to-dos"
AI->>M: Writes nested to-dos and diagrams
M->>You: Shows it at a glance as a diagram
You->>M: Edit and check off
M->>AI: The AI reads your edits and checks
AI->>M: Adds the next tasks and diagrams
```
> [!IMPORTANT]
> ### Get started → [User guide](https://github.com/nhorie-canal/mmm_mcp/blob/main/docs/guide.md)
>
> From preparing the app to connecting your AI, the recommended instructions and everyday use, in order.
## Privacy
- It reads and writes your maps as you, only when your AI asks. The same access control as the app applies
- Map contents are not stored on the server or written to logs. The server keeps only connection grants (hashed) and daily call counts (deleted after 3 days)
- The days you wrote (dates and a count only) and your device's time zone are kept in a part of your account only you can read. The former is used only to count users
- The first time a map's contents are read, the AI asks for your permission
- You can disconnect from your AI's settings. Deleting your account removes the grants. Content passed to your AI is subject to its provider's terms
See [Privacy](https://info.spheres.jp/en/apps/matryoshka-mind-map/#privacy) for details.
TDQS
Scored across 14 tools
Most tools map cleanly to a distinct resource+action: list_maps/list_elements for reads, create_map/create_document for creates, edit_element/move_element/set_schedule/update_task_status for distinct update kinds, and delete_elements/sign_in are unambiguous. The main soft spot is import_markdown_context vs auto_structure_thought, which both write hierarchical content into an existing map; the descriptions differentiate the content source, but an agent could still plausibly pick the wrong one.
Thirteen of fourteen tools follow a clear verb_noun snake_case pattern (list_elements, create_document, update_task_status, delete_elements, move_element, etc.). Only sign_in (verb+particle) and the more idiosyncratic auto_structure_thought break the pattern, which is a minor deviation rather than a different convention.
Fourteen tools sit comfortably in the well-scoped range, and each covers a real capability of the domain: map listing/creation, element CRUD-ish operations, documents, scheduling, task status, moving, and auth. Nothing looks redundant enough to be padding.
The element lifecycle is essentially covered (read via list_elements, add via auto_structure_thought/import_markdown_context, edit/move/schedule/check via the update tools, remove via delete_elements), and documents have read/create/update. Gaps remain at the container level: there is no map deletion, rename, or share management tool, and no explicit document-delete (agents must infer it via delete_elements).