You saved something worth coming back to. Give your assistant the instructions to bring it into your collection, check what arrived, and pick up where you left off.
Two skill packs. One private ledger. Built for people who collect ideas before they know what they will write.
The MCP connection gives your assistant tools. These skills teach it the workflow: confirm the source account, capture saved posts, import in batches, and report what is complete or still missing.
Bring in what is new. Refresh LinkedIn saved posts and X bookmarks through a supported, signed-in browser.
Keep your place. Resume from recorded scan progress and completed imports after an interruption.
Know what arrived. See imported, updated, unchanged and rejected counts, plus complete or partial source coverage.
A skill is an instruction pack, not a background service. Refresh runs when you ask, while your assistant has access to the browser or an export. Signing in alone does not import bookmarks.
Choose your assistant.
Each ZIP includes the skill and its workflow references. Download, unzip, and read the files before installing. Nothing in the download contains account data or credentials.
If that server name already exists, check its URL before replacing anything. Choose the same Google account as your ledger and review the connection approval.
Start a new Codex session, select $codex-saved-ledger, then ask it to check the connection and show your ledger counts.
Move the extracted claude-saved-ledger folder into ~/.claude/skills/.
Run this command in your terminal:
claude mcp add --transport http --scope user saved-ledger https://savedledger.com/mcp
If already configured, check its URL first. In Claude Code, run /mcp, choose saved-ledger and Authenticate. Use the same Google account as your ledger and review the connection approval.
Start a new session and use /claude-saved-ledger. Ask it to check the connection and show your ledger counts.
Select and copy a prompt into your connected assistant. Change the source or topic to fit your collection.
Read the engagement and archive workflow. Open a category’s Engagement queue to review drafts, record outcomes and reopen archived work. Publishing and following happen on the original platform; archive and source unsaving are separate actions.
A refresh you can trust.
“Complete” needs evidence.
A quiet scroll or a timeout is not proof. The skill requires an explicit end or independent source total. Otherwise it reports partial coverage.
Your organisation stays yours.
Repeated imports deduplicate posts. Longer captures can improve stored text; confirmed categories survive. Shorter edits do not overwrite stored text, so this is not a live mirror of every edit.
Sync keeps source saves.
Cleanup is a separate, explicitly requested workflow. It checks the archive before unsaving and records outcomes. A URL-only export is not enough to authorise removal.
Private still means a data flow.
Authorised assistants can read returned post data. Browser capture through tool arguments passes text through your assistant conversation; direct uploads avoid that step. Server-side AI suggestions use the consent setting in your ledger.
Make it a writing habit.
Before your next research or writing session, ask for a refresh. Review the coverage report, then search your collection. For a scheduled routine, use an assistant that supports scheduling and can access your signed-in browser; these downloads do not enable unattended syncing.
No browser access, expired login, or a stalled import?
Sign in again from your assistant if its MCP grant has expired. If a browser source asks for authentication, complete that yourself. You can import capture JSON or a LinkedIn Saved Items CSV/ZIP instead; LinkedIn exports may contain only links and dates. X account archives are not supported.
Finished imports survive interruption. Expired upload staging needs a new sync and recapture. Ask for the resume workflow instead of assuming the last batch finished.