Morion
ローカルmacOSノートブックであり、MCPサーバーでもあります。Cl
ドキュメント
Connect Claude Code, Codex, Cursor and other agents to your notes and tasks.
Morion runs a local MCP server next to your agents. Claude Code, Codex, Gemini Antigravity, Cursor, Cline, Zed and any MCP client get the same tools and the same data. A Claude session and a Codex session can work on the same board at the same time, on your machine, with every write attributed.
local serverany clientper-folder permissions
Settings · MCP Server
Settings · MCP Server
General
Limits
Usage
Chat Settings
Chat Memory
Integrations
MCP Server
Skills
Logs
MCP Server
Enabled
AI apps can read and write your notebook.
Read
Search, list and read notes, folders and tasks; Ask Mo.
Update
Edit notes, move cards, claim tasks, rename folders.
Workflows
Auto-code pipelines and the folder, project and workspace auto-code settings.
| Actor | Calls | Last seen |
|---|---|---|
| mcp:claude-code | 1,204 | 2m ago |
| mcp:codex | 318 | 41m ago |
| mcp:cursor | 57 | Yesterday |
The master switch, the five tool categories and the clients that called tools recently.
How it works
1
Connect a client in one click
Morion detects installed clients, writes the MCP entry and backs up the config file first. Anything else gets a snippet to paste.
2
Decide what it may do
Five categories under the master switch: Read, Create, Update, Delete, Workflows. Then per folder: visible, create, update, delete. Per note for the exceptions.
3
Run agents in parallel
Several sessions read and write the same notes and tickets. Claims are atomic, moves are logged, nothing is synced through a server.
4
See who did what
Every write records the actor. The Connected clients table shows calls and last seen per client; the audit log is readable in the app and over MCP.
In detail
Skills for the agents
Installed from Settings, Skills: how to use the board, how to ask Mo, how to edit workflows safely. Your CLAUDE.md keeps the rules; Morion keeps the memory.
Hidden means hidden
A folder hidden from AI is invisible to every client and to Mo, and never appears in a work packet or an answer. Archived notes are hidden from MCP too.
Actors, not just "AI"
mcp:claude-code, mcp:codex, mcp:cursor: the client names itself, and the name is on every comment, move and edit.
Local only
MCP over stdio on your machine. No HTTP endpoint, no cloud relay, nothing to expose.
Mo over MCP too
The Mo tools are part of the same surface, so an agent can ask Mo, take a work packet, or leave you a question from its own session.
What it is not
Not a code tool
The tools touch notes, tasks, folders, tags and workflows. They do not read your repo or run commands; your agent does that with its own tools.
What is MCP
Model Context Protocol is an open standard that lets an AI assistant call tools on a server. Instead of pasting context into every chat, your assistant connects to Morion and pulls what it needs on its own, and writes back.
The raw tools give any client the notebook and the board: search, read, create, claim, move, comment. The mo_* tools give it the engine: a work packet with the ticket and its context, an answer with a citation and a status on every claim, a conversational lookup.
Every write is logged with the actor: mcp:claude-code, mcp:codex, morion-concierge for Mo, user for you.
Topologystdio · local
Morion
one file on your disk
MCP · stdio
Claude
Codex
Gemini
Cursor
Cline
Zed
Any MCP client
same protocol, same tools
Every tool, by what it touches. 57 tools · v1.6.10
Every tool is available to every connected client, subject to the folder and note permissions you set. In the app the tools sit in five categories you can switch off one by one: Read (20), Create (9), Update (11), Delete (5) and Workflows (12). Below they are grouped by what they touch; the Workflows table is the whole Workflows category.
On the Free plan, external MCP clients get 1,000 calls a month; Pro has no limit. Calls from Mo and Auto-code, and failed calls, never count.
Mo (11 tools)
| mo_take_task | Primary pickup: ticket body, relevant comments, project brief and the notes needed before starting, picked at request time, in one call; optionally claims the ticket |
|---|---|
| mo_ask | Question answering over retrieved evidence: every claim cited and status-tagged, conflicts and gaps reported; scope by project, folder or kanban statuses |
| mo_find | Conversational lookup ("find that ticket", "similar to this note") with a per-hit reason |
| mo_search | Hybrid keyword and semantic search across notes; each hit comes with a body excerpt; no synthesis |
| mo_list_projects | Project directory: each project’s folders, their roles and the project brief |
| mo_list_tasks_meta | Task metadata without bodies: id, title, status and folder; filter by folder or text |
| mo_resolve_task | Resolve a task reference to its record: body, folder, status, tags, recent comments and history |
| mo_remember | Store a workspace-wide fact; conflicts are surfaced, not overwritten |
| mo_forget | Remove workspace memory by pattern or entirely |
| mo_request_human | Durable escalation: a comment on the task or a backlog card |
| mo_check_workflow | Pre-flight gate for an intended action: allow, deny or ask the user |
Mo tools work only in folders where AI access is on, and never in a folder hidden from AI. What Mo writes is advisory: it never corrects your own edits.
Notes (16 tools)
| notes_search | Hybrid keyword + semantic search across all notes |
|---|---|
| notes_list | List notes with folder, tag and pinned filters and pagination |
| notes_get | Fetch a note by ID with full body and tags |
| notes_create | Create a note with optional folder, tags and status |
| notes_update | Partial update; only provided fields change |
| notes_append | Append text to a note without rewriting it |
| notes_delete | Soft-delete to trash |
| notes_duplicate | Clone a note with a fresh ID |
| notes_move | Move a note to another folder |
| notes_recent | Last N notes by update time |
| notes_list_activity | Unified comments + audit feed, cursor-paginated |
| notes_add_comment | Post a comment, optionally as a reply |
| notes_update_comment | Edit your own comment |
| notes_delete_comment | Remove your own comment, audit-logged |
| notes_list_attachments | Attachment metadata only, cheap to call |
| notes_get_attachment | Returns an attached image so the agent can see it |
Tasks (4 tools)
| tasks_list | Tasks in a kanban folder, filtered by status and date |
|---|---|
| tasks_claim | Atomically claim a todo or reopened task into doing; no races between agents |
| tasks_move | Move a card between columns with a message, or reorder within a column |
| tasks_history | Full status-transition history: who moved what, when |
Seven columns: note · backlog · todo · doing · review · reopened · done. A task is a note with a status: same search, same folders, same permissions.
Workflows (12 tools)
| workflows_list | List Auto-code workflows for a project |
|---|---|
| workflows_create | Create a workflow definition |
| workflows_update | Update a definition |
| workflows_copy | Copy a workflow |
| workflows_delete | Delete a workflow |
| workflows_validate | Validate a definition before installing it |
| workflows_environment | Report the agents and settings available to workflows |
| mo_build_workflow | Author an Auto-code workflow definition from a description |
| folders_get_auto_code | Read a folder's Auto-code settings |
| folders_set_auto_code | Turn Auto-code on or off for a folder and choose its workflow |
| projects_set_auto_code | Apply one Auto-code setup to every folder of a project: Mo access, the repository, the workflow, concurrency and auto-merge |
| auto_code_workspace_settings | Read or change the workspace-wide Auto-code settings |
The morion-workflows skill teaches an agent to author and install these safely.
Folders (9 tools)
| folders_list | All folders, ordered |
|---|---|
| folders_create | Create a folder |
| folders_rename | Rename |
| folders_delete | Delete |
| folders_duplicate | Clone with notes |
| folders_move | Move in the sidebar |
| folders_reorder | Reorder by ID list |
| folders_set_project | Move a folder to another project |
| folders_set_view_mode | List or kanban view |
Tags (4) and audit (1)
| tags_list | List tags |
|---|---|
| tags_create | Create a tag |
| tags_update | Rename or recolor |
| tags_delete | Delete; notes keep their content |
| audit_recent | Last N audit entries: what each client changed |
Every MCP write is logged with the client name.
Go deeper
[
Integrations
One-click setup per client, and the model sources Mo and Auto-code run on.
](https://morion.ai/features/integrations)[
Three ways to orchestrate
Subagents on the board, Auto-code, or both.
](https://morion.ai/use-cases/three-ways-to-orchestrate)
More features
[
Notes and documents
Markdown notes on your disk that your agents read and write.
](https://morion.ai/features/notes)[
Kanban tasks
A private board where agents claim tickets and you review.
](https://morion.ai/features/kanban)[
Integrations
One-click setup for Claude Code, Codex, Cursor and more; the model sources Mo and Auto-code run on.
](https://morion.ai/features/integrations)[
Permissions
Per-folder and per-note access for agents, tool categories, audit log.
](https://morion.ai/features/permissions)[
Smart retrieval
Mo picks the notes each task needs at the moment an agent asks.
](https://morion.ai/features/indexing)[
Mo Assistant
Ask what was decided and get the source note. Mo works the notebook while your agents work the code.
](https://morion.ai/features/mo)[
Auto-code
Move a ticket to Todo; Claude writes, Codex reviews, you merge with one click.