No context path between claude.ai and Claude Code

Status Open
Reported on v2.1.233
Maintainer reply None cached
Activity 1 comment · opened Aug 16, 2026

Environment: Windows 11 Pro 26200, Claude Code 2.1.233, Max plan. One
account, one login, both products signed in as the same user.

Summary

claude.ai and Claude Code hold separate, unconnected memory stores. Nothing
flows between them in either direction, despite being the same product family on
the same authenticated account.

  • claude.ai projects have a Memory panel and project Context files.
  • Claude Code has auto memory at ~/.claude/projects/<project>/memory/.
  • There is no supported path from either to the other.

The bridge already exists, and carries the wrong thing

With remoteControlAtStartup, every Claude Code session registers itself to
claude.ai/code and can be driven from the web app or the mobile app. That is a
live, authenticated, bidirectional channel between the two products, working
today.

It carries session control. It does not carry memory, project context, or user
preferences. The plumbing is in place; only the payload is missing.

What users do instead

The only method that actually works for getting claude.ai project memory into a
Claude Code session is to point Claude Code's own browser automation at
claude.ai and read the rendered page text. It works — the Memory panel, the
conversation list and full conversations all read as ordinary page text.

Reading your own data, from one product of the same company, on the same login,
by scraping the other product's DOM, should not be the supported route.

It also fails partway: project Context files cannot be read this way,
because the Context folder does not expand into the accessibility tree. Those
still have to be copied out by hand.

Measured cost, one user, one week

  • Fifteen memory entries transcribed by hand from claude.ai into Claude Code,

because no export path exists.

  • A false rule — "claude.ai memory cannot be read" — sat in the user's global

CLAUDE.md for four days, because the true answer was neither documented nor
discoverable.

  • Project Context files remain unreadable and must still be pasted manually.

Requested

  1. **Read access from Claude Code to the signed-in account's claude.ai project

memory and project Context files**, opt-in and scoped per project.

  1. Ideally bidirectional: memory Claude Code writes for a project should be

visible in that project on claude.ai, and vice versa. One account, one set of
things Claude knows.

  1. Failing either, a supported export/import path that is not browser

automation against your own web app.

Related

  • anthropics/claude-code#87027 — account-level sync for user config and auto

memory across machines. Related but distinct: that one is machine-to-machine,
this one is product-to-product.

  • anthropics/claude-code#86999 — installer and claude doctor defects.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗