Remote MCP made your Tana Outliner graph reachable from anywhere. The instruction files that teach an agent how to use it can travel the same way. Building on our Remote MCP sessions, in this Systems Lab we explore a portable knowledge core: structured knowledge in Tana Outliner, reusable instructions in GitHub, so AI tools like Claude and ChatGPT can work with your system wherever you are, including from a phone at the gym.
Mark walks through his workout and nutrition examples, connects Claude to both Outliner and GitHub, and explains how skills add workflow knowledge to MCP's tools. MCP gives an agent the tools to act in Outliner. It does not tell the agent what your system means, which is the job of the instruction files. The example repo from the session is public: portable-knowledge-core.
We then use guided discovery to build a first version of a reading system in Outliner, with captures, reading ideas, and quizzes. Along the way we discuss content capture, defining useful outcomes, and turning collected information into knowledge you can put into practice.
We'll look at:
- Keeping instructions in GitHub so the same workflow follows you between AI tools
- The difference between MCP's tools and the skills that tell an agent how to combine them
- Using guided discovery to design a system, and what makes an outcome worth building toward
What is a portable knowledge core? [3:21]
Remote MCP made the graph portable. The instruction files can be too. Put them in a GitHub repo, connect GitHub to your AI client, and the same understanding of your system travels with you. The demo repo holds two working skills: create-tana-system for starting from nothing, and discover-tana-system for reading the supertags and fields you already have and writing them up as a skill file.
Workout and nutrition examples [5:05]
A training plan lives in Outliner with the exercises and the previous session's reps and sets. Asking ChatGPT for the next workout pulls both, voice mode reads the session out at the gym, and "log all this to Tana" hands off to the logging skill, which is what makes the result land as structure rather than a paragraph. The nutrition example starts from nothing more than "I want to track what I eat" and proposes supertags, fields, and a search node for food items. Because the data is in Outliner, the same chat can then chart calories per day.
MCP tools vs. skills [13:03]
MCP is fundamentally a list of tools: list workspaces, get a supertag's schema, read a node, set a field value. Every manual action in Outliner has a matching tool. What the tool list does not carry is which combination to use, or why. It knows how to create a field called calories; it does not know what your system means by one. That is the job of the AGENTS.md or CLAUDE.md a client reads at the start of a session, and of the skills beside it.
Guided discovery: designing a Tana system [18:16]
The guided discovery skill is a written version of a coaching conversation: what matters to you, what you are logging, how you want to see it. MCP has made building the system cheap and largely automatic, which leaves knowing what you are trying to achieve as the real work.
Building a structured reading system [23:48]
The live build starts from the read-it-later reflex that fills a backlog nobody returns to. The requirements come out clearly: no summaries, because they are a lossy version of an article whose value often sits in one sentence; the full text in Outliner, so the record survives the link. Success is not a bigger collection but returning to the big ideas and being tested on them, with you as the filter for what counts as important.
Making the workflow reusable across AI tools [54:15]
The first pass produces a reading capture tag with fields for the source URL and raw text, plus a quiz tag with a hide condition on the answer so a question can be asked before the answer shows. Writing that structure into a system file and skills in the repo, so any client with access can run it, is the next step and is described rather than built: the session ran out of time before doing it.