Codex takes a task and goes away with it, which is exactly when the reasons behind the code matter most and are least available. Keep them as a structure on your own disk and add it to Codex as an MCP server in one TOML table, and the agent can read the constraints before it starts and write down what it decided when it’s done. Free on every plan.
Add a table named mcp_servers.filamental to Codex’s config.toml, running node against the bundled server with your space folder passed after a --vault argument:
[mcp_servers.filamental]
command = "node"
args = ["--no-warnings", "/absolute/path/to/filamental-mcp/dist/index.js", "--vault", "/absolute/path/to/your/vault"]Or let Codex write it for you, with everything after the two dashes being the command it should run:
codex mcp add filamental -- node --no-warnings /absolute/path/to/filamental-mcp/dist/index.js --vault /absolute/path/to/your/vaultThe file lives in the .codex folder of your home directory, and a project can carry its own .codex/config.toml, which Codex reads for projects you’ve marked as trusted. The Codex CLI, its IDE extension and the ChatGPT desktop app all read the same configuration, so you set it up once. Copy both absolute paths from Settings, then AI, then Other AI apps inside Filamental rather than typing them, since the server ships with the application and lands somewhere platform-specific. Filamental shows them in JSON, and the two paths are all you need from it.
On Windows, mind the backslashes. TOML reads a backslash inside double quotes as the start of an escape, so write the paths with forward slashes, which Node is happy with, or put them in single quotes, which TOML takes literally.
Delegated work tends to go wrong in one particular way. The agent does exactly what was asked and undoes something nobody wrote down, a constraint agreed in a meeting or a boundary drawn for a reason that left with the person who drew it.
In a space those reasons are typed connections between named things, so blocks, depends on, supersedes and was rejected because are separate relationships with a direction each. Codex can read them before it starts, follow them while it works, and ask for the route between two things when it wants to know whether they’re related at all.
It writes as well as reads, so the last step of a task can be recording what was decided and why, in the place the next task will look first.
A short briefing arrives when the connection opens, covering what nodes, categories and connectors are and the conventions that keep a structure readable, and the full skill guide and file format specification are pulled on demand only when a task needs them.
Both are published on this website in full, word for word as they are sent: the skill file and the file format reference. Nothing else runs underneath them. The server is a local process with no network access of its own, and Filamental has no account, no sign-in and no telemetry of any kind.
One prerequisite, and it is the only one that catches people out. The server reads the index Filamental builds and never builds one itself, so the space has to have been opened in the application at least once on that machine. A folder it has not seen stops with a message saying exactly that. If reading a folder somebody sent you is what you want, point the AI at the files directly instead, since a space is plain Markdown and no index is involved.
You need Node.js 22 or later and Filamental v0.2.4 or later, on Windows, macOS or Linux. The server ships inside the application, so there is nothing separate to install, and it is also on npm as filamental-mcp for a machine without Filamental on it.
The connection is free on the Personal plan, which is free permanently with no account and no card, and it is not metered, credited or seat-limited. The paid tier is $120 a year, or $12 a month, and covers handing work over beyond the one live link that’s free.
Add a table named mcp_servers.filamental to config.toml in the .codex folder of your home directory, with command set to node and args holding the bundled server's path, then --vault and your space folder. Or run codex mcp add filamental, then two dashes, then that same command, and Codex writes the entry itself. Copy both absolute paths from Settings, then AI, then Other AI apps inside Filamental.
No. According to OpenAI's documentation the Codex CLI, the IDE extension and the ChatGPT desktop app read the same config.toml, so one entry reaches all three. A project can also carry its own .codex/config.toml, which Codex reads for projects you have marked as trusted.
Usually because of the paths. TOML treats a backslash inside double quotes as the start of an escape sequence, so a Windows path written the usual way breaks the file. Write the paths with forward slashes, which Node accepts on Windows, or put them in single quotes, which TOML reads literally.
The reasons. Codex can read what the code does, but not why a boundary sits where it does, which constraint ruled out the obvious approach, or which decision a change would quietly reverse. In a space those are typed connections between named things, so the agent can read them before it starts a task, follow them while it works, and record what it decided when it finishes.
No, because there is nothing of ours in the path. The server is a local process talking to Markdown files in a folder on your own disk, and whatever leaves the machine is what Codex sends to OpenAI, exactly as it does for the rest of your work. Filamental has no account, no sign-in and no telemetry of any kind.
Nothing. It is free on the Personal plan, which is free permanently with no account and no card, and it is not metered, credited or seat-limited. The paid tier, at $120 a year or $12 a month, covers handing work over beyond the one live link that's free, such as more links, updating a link already sent, the HTML download and presenting.
Ask Codex to record the decision and the constraints behind it in your space when it finishes. The next task, and the next person, start from there instead of from the diff.