Your team uses Notion or Obsidian as a second brain. Where should the strategy live?

Keep the notes where they are. A second brain is built to capture and distill what you read, hear and think, for your future self. A strategy is something else: a few commitments a committee owns, linked to each other, with an owner, a status and a review date. Put those in a bounded, structured record, and let your AI agent work across both.

The agent is the bridge, not a connector. Notion runs an official MCP server, and an Obsidian vault is a folder of Markdown files that an agent on your machine can read. Hoshin exposes the X-Matrix through its own MCP server. Connected to both, the agent turns a meeting note into an idea for the strategy inbox, or writes the review brief back into your notes. Hoshin never reads your Notion or your vault: only your agent does, with your rights.

What a second brain is for

The best-known method is Tiago Forte's, in four steps: capture information from the outside world, organize it so you can find it, distill it to the best ideas, express them in your own work. It is a personal knowledge system, and it works best when it is generous: you keep more than you will use.

The two tools people name most often take that idea in different directions:

  • Notion: a shared workspace of pages and databases. Its hosted MCP server lets an assistant search, read, create and update content, and Notion states that the assistant can only reach what the authorizing member can access.
  • Obsidian: notes as plain Markdown files in a folder on your own disk, the vault. Any tool that reads files can read them, which is why an agent running on your machine (Claude Code, Cursor) can read it. With the Obsidian 1.12 installer or later, the Obsidian CLI also lets a script create, read, search and append to notes while the app is running.

Why the strategy fits badly in a second brain

A second brain is unbounded on purpose. A strategy has to be bounded: three to five long-term objectives, a handful of annual ones, the initiatives that serve them and the indicators that prove them. Mixed into thousands of notes, it becomes one page among others, and it is stale the week after the offsite.

The links are the other problem. A wikilink or a Notion relation says that two pages are connected. It does not say that this initiative serves that annual objective, that this KPI measures that initiative, or that a correlation only exists between adjacent quadrants of the X-Matrix. Without those rules written as data, the agent infers the structure from prose at every session, and two agents infer two different strategies.

You can build an X-Matrix in Notion with four databases and relations. What you then rebuild by hand: the correlation corners, the completeness score, the cascade from group to site, write rights per matrix, and the review cadence that keeps statuses current. In Obsidian the obstacle is simpler: a vault is personal, and a strategy belongs to a committee.

How one agent works across both

Connect your assistant to Hoshin (claude.ai, Claude Desktop or Cursor, through OAuth) and to Notion through its MCP server, or give it access to your vault. Then ask, in plain words. Four patterns, using tools that exist today on our side:

  • From the meeting note to the strategy inbox: the agent reads the note, spots what touches an objective, and files it with add_idea (source team_meeting, linked to the element concerned). The committee sorts the inbox; the agent decides nothing.
  • From the matrix back to your notes: the agent reads the matrix and its score (get_matrix_data, get_matrix_completeness) and writes the review brief as a Notion page or a note in the vault, where the team already reads.
  • A link from each element to its working page: an element's external reference can hold the URL of the Notion page where the work is followed. The agent that updates a status knows where to look.
  • The weekly sync with Notion as the execution tool: if your projects are tracked in a Notion database, start the weekly_sync prompt with execution_tool set to Notion. The agent reads the initiatives, checks their progress in Notion, updates statuses and prepares the agenda.

Obsidian users and the local mode

People choose Obsidian because the files stay on their machine. Hoshin has the same option: the open-source MCP server runs in a local mode that keeps the matrix as JSON files in a folder you choose, with no account and no server of ours. Your agent then reads your vault and your matrix on the same machine, and nothing reaches us; what your agent's model sees is between you and its provider.

The local mode is for one person or a consultant preparing a client matrix. As soon as a committee shares the matrix, it needs the hosted app or your own deployment: rights per matrix and the audit log live there.

What we have tested, and what we have not

The API routes our tools call are tested end to end. The weekly_sync prompt is a protocol the agent follows, and its result depends on the agent. Notion's server and Obsidian's files are documented by their makers, and we cite their documentation below. We have not run the four patterns end to end with a real Notion workspace or a real vault. They rely only on standard MCP, so any agent that holds both connections can compose them; how well depends on the agent and on how your notes are written. If you try it, tell us what broke.

Frequently asked questions

Does Hoshin integrate with Notion?

Not through a connector. Your agent connects to Notion through Notion's own MCP server and to Hoshin through ours, then carries information between them when you ask. Hoshin never holds a Notion token and never reads your workspace.

Does Hoshin work with Obsidian?

Through your agent. An Obsidian vault is a folder of Markdown files, so an agent running on your machine (Claude Code, Cursor, or a script using the Obsidian CLI) can read and write notes, while it reads and writes the strategy through Hoshin's MCP server.

Can I build a Hoshin X-Matrix in Notion instead?

Yes, with four databases and relations. You then maintain by hand what a dedicated model computes: correlations restricted to adjacent quadrants, the completeness score, the cascade between levels, per-matrix write rights and the review cadence. It works while one person keeps it up.

What does Hoshin see of my notes?

Only what your agent writes through Hoshin's tools, for example the title and description of an idea it files in the inbox. Your notes themselves stay in Notion or in your vault.

Sources