Get started with Living
Get going in four steps
Public sign-up and connections are still in development. The steps below preview public Beta setup. Parts of the MCP and Skills flow have been verified in the local Beta.
Create a Project
Coming SoonSign in to Living, create a Project, then enter its name and choose a language.
Setup screenshots will be available when public access opens.
Issue an MCP key
Coming SoonIssue an MCP key in the Project’s connection settings, store it securely, and keep plaintext keys out of Git.
Add the Living MCP server
Coming SoonWhen public access opens, get the server URL and authentication instructions from the Project’s connection settings.
Claude Code
- From the project folder, use the Claude Code CLI to add a remote server with
claude mcp addand the HTTP transport. Use local scope for your own connection. - Use the URL and any authentication options provided in the Project’s connection settings. The CLI supports an Authorization header for bearer-token authentication.
- Open Claude Code in the project and run
/mcp. Confirm that Living appears as connected.
OS verification details
- macOS
- Verified in local Beta
- Windows
- Not verified
- Linux
- Not verified
The author tested the local Beta on macOS through an adapter on the same machine. Codex MCP calls were verified through a bridge. The public remote connection is pending on all platforms.
Codex
- Open Settings → MCP servers → Add server. Choose Streamable HTTP and enter the Living server name and URL.
- Follow the Project’s authentication instructions. If it uses a bearer token, enter the name of a token environment variable available to the app.
- Save and select Restart. In the composer, use
/mcpto check that Living is connected.
Advanced: manual TOML configuration
For manual configuration, add the server to your user-level
~/.codex/config.tomlusing the URL and authentication settings provided for your Project.OS verification details
- macOS
- Partly verified in local Beta
- Windows
- Not verified
- Linux
- Not verified
The author tested the local Beta on macOS through an adapter on the same machine. Codex MCP calls were verified through a bridge. The public remote connection is pending on all platforms.
- From the project folder, use the Claude Code CLI to add a remote server with
Initialize the project
Public setup requires a Project and a working MCP connection. From the connected project, send this prompt to your coding agent.
Send to your agentPublic access is pendingInitialize Living for this project
Your agent gets compatible Skills over MCP, installs them in the project, initializes
.living/and verifies the setup.The initialization flow was verified in the author’s local Beta, where engineering data stays local. Automatic Skill loading in the Codex desktop app is still unverified.
Prompt library
Use these prompts with Living Skills once your project is initialized.
- Initialize the projectGo to step 4
Record what you builtTurn finished work and its actual result into an outcome record.
Turn the work we just finished into Living engineering records. State the actual result and the evidence, and keep only what will matter later. Save them locally. Don’t sync.
Requires · Project initialized
Record a decisionKeep the approach, reasons and alternatives — proposed vs. adopted.
Record the design discussion we just had as a Living decision: the approach, the reasons, the alternatives we considered, the evidence and the open questions. Anything I haven’t explicitly approved stays as proposed.
Requires · Project initialized
Map the architectureDescribe the current architecture from real code and docs, with sources.
Map this project’s current architecture from the actual code, config, docs and tests, as sourced Living engineering knowledge. Keep observed, inferred and verified apart, and leave anything uncertain as unknown. Don’t sync yet.
Requires · Project initialized
Review the historyTrace phases and key decisions; you decide what gets recorded.
Review this project’s engineering history: phases, key decisions, the evidence we have and what’s still unresolved. Cite a source for each point and mark anything without one as unknown. Show me before writing any records.
Requires · Project initialized
Validate and syncValidate and sync the confirmed scope, then check the receipt.
Sync the new and changed Living records to Living. First list what you plan to submit and wait for my confirmation, validate it, then sync and tell me what the receipt says.
Requires · Only when you want to sync
Data defaults in the local Beta
- Engineering records are saved locally first.
- Auto-sync is off by default.
- Submitting to Living requires your explicit approval.
Data and security defaults
- You set the sync scope
- Before syncing, the agent lists what it will submit for you to choose or confirm.
- A suggestion isn’t approval
- The agent can only record a decision as proposed; adopted needs your explicit word.
- Nothing gets overwritten
- Skills or project files you already have, or changed, are left alone during setup.
- Credentials stay out
- The MCP key never goes into
.living/, Git or a public page. - Separate settings
- The Project’s language, your agent’s MCP setup and the local
.living/are set independently.
FAQ
Can I connect on my own now?
Not yet. Sign-up, Projects, MCP keys and remote connections are in preparation, and Living itself has not been released. When they open, this page will switch to steps you can run directly.
What’s the difference between a Project ID and an MCP key?
The Project ID identifies your project and is stored in .living/project.json. Manage the MCP key as a credential using the options your agent supports. Keep plaintext keys out of .living/, Git and public pages.
What does initializing add to my project?
Setup checks the existing Project, verifies the file hashes for the pinned Skills version and binds the project folder to that Project. A .living/ folder and the installed Skills. sync/ belongs to one Living instance and usually isn’t worth committing.
.living/project.jsonproject binding, no credentialsrecords/observations, verifications, outcomes, issues, decisions, hypothesesknowledge/engineering knowledge with sourcesevidence/files that records refer tosync/sync state and every receipt
What happens when I sync?
In the local Beta, every sync request follows the same rule. The agent must get your explicit approval of the scope before it submits anything. The agent validates what it is about to submit, then sends it as one batch: written completely or not at all. Living returns a receipt saying, item by item, what was created, updated, unchanged or rejected, kept in .living/sync/receipts/. If something is rejected, fix what the receipt reports and sync again.
Does Living collect my agent sessions?
The local Beta does not collect agent conversations. It syncs records and knowledge at your explicit request. Living Web in the local Beta does not access the project directory directly. The public Beta’s privacy policy will be published when public access opens.
My agent doesn’t see Living’s tools
MCP configuration takes effect in a new session, so start a fresh one. Newly installed Skills are also loaded when a session starts.
Initialization stopped
Usually a Skill with the same name was changed locally or wasn’t installed by Living, or a downloaded file failed its hash check. Nothing is written in that case; deal with the reason the agent reports, then initialize again.
Learn more
Living Skills
The Living Skills source is publicly available at the version below. You can view or clone it without a Living account.
Some Skills, including living-init, require Living’s MCP Bootstrap and Contract. A complete standalone workflow has yet to be verified. Consult the repository for documented capabilities. This version has no open-source license.
git clone --depth 1 --branch v0.4.0-rc.1 https://github.com/kenkokuLife/living-skills.gitThis command fetches the source files.