[FEATURE] Enable Claude Code to access chat history in Claude App

I've also been working on a project in both Claude and Claude Code. I'm getting to the point where I need access to the Claude project chats in order to add documentation to the repo I use with Claude Code.

Claude linked to this issue when I asked about sharing history between the two.

Adding my +1 — I use Claude across VS Code agent, claude.ai browser, and Cursor for Laravel development. Cross-surface history would be a game changer for professional workflows.

I'd then be able to export those web/desktop conversations to markdown since it's not currently possible contrary to perplexity.ai for example.

+1

Suggested integration: let me expose claude.ai (either conversation history directly, or a conversation that can access that history) as an MCP server that my Claude Code can access. Seems clean and to reuse existing machinery?

Adding perspective from 50+ Cowork sessions running a retail fireworks business entirely through Claude collaboration.

The silo problem extends beyond Code ↔ Chat. Chat, Cowork, and Code Tab share zero context with each other. I design architecture in Chat, then switch to Cowork to execute — and start from scratch every time. The manual context-carrying defeats the purpose of an intelligent agent.

What makes this worse in practice:

• Cowork Projects memory (March 2026) and cross-session memory partially help, but only within a single surface. Cross-surface continuity is still absent.
• - Cross-machine sync is nonexistent. I work on an M1 MacBook (primary) and M3 iMac — context built on one machine doesn't follow me to the other.
• - - I've built an entire custom skill and context-file system (CLAUDE.md, _claude-context/, session-reload skills) to manually solve what should be a platform feature. The fact that power users are building their own persistence layers proves the gap is real.

The cost isn't just friction — it's lost work. Context compaction in Cowork already silently discards information mid-session. When you combine that with zero continuity across surfaces, users lose work in two directions: within a session (compaction) and between sessions/surfaces (no shared memory).

This is the single highest-impact improvement Anthropic could make for users who treat Claude as a daily working tool rather than a one-off chat assistant.

+1

+1

Also, it would be great to have the ability in Claude for VS Code to start a chat that's stored in my personal history which would basically allow me to have the same experience that we have now when we use Claude Desktop, Claude for mobile, and Claude Web.

+1

Use case: continuing claude.ai conversations in Claude Code (CLI/VS Code)
I regularly use claude.ai for planning, research, and architectural discussions — then switch to Claude Code to implement. Every time I have to manually copy-paste a summary because Claude Code has zero awareness of what was discussed in claude.ai, even when authenticated with the same account.
This fragmentation is genuinely frustrating — and it's especially hard to accept coming from a product that actively markets itself as a multi-agent system. You have claude.ai, Claude Code CLI, the VS Code extension, the Chrome extension, the mobile app — all under the same account, all supposedly part of the same ecosystem, and none of them share conversation context. That's not a multi-agent system, that's a collection of isolated chatbots wearing the same brand.
The context already exists. The user is authenticated. There is no technical reason these should be completely isolated silos.
Minimum viable fix: read-only access from Claude Code to claude.ai conversation history for the authenticated user — same account, opt-in, no sync needed in the other direction.
The current workaround (manually summarizing and pasting context) defeats the entire purpose of persistent memory and cross-session continuity that Anthropic is actively building and marketing.

the fragmentation is frustrating, claude.ai chat long planning session, won't flow over into claude co-work, or claude.ai - forcing me to find alternative ways to manage the collaboration, setting up .md files, then wasting tokens ingesting them into a new claude code session, and if you use skills in claude.ai, you need a way in import those skills as well. it interrupts the thinking

As a terrible workaround, you can export all data from the desktop claude.ai website. (24 hour turnaround to honour the request.) The resulting zip does contain everything, and a utility like https://github.com/risaacr/claude-chats.git will extract it into a local filesystem with some sane deduping.

But come on Anthropic - you are delivering features so fast, and an MCP to claude.ai for Claude Code seems like quite a small one?

This is one slice of a larger pattern — surfaces operating as silos despite sharing one authenticated user. Memory, files, skills/connectors, and chat history all need to converge. Filed a framing-level issue at #55842 that proposes treating Cowork and chat as two surfaces over one user state.

This is an excellent feature request that identifies a real structural gap. The architecture you want — claude.ai conversation → Claude Code implementation with shared context — is fundamentally a memory problem, not a chat sync problem.

The key insight from your proposal: the context Claude needs isn't the raw chat transcript. It's the decisions, constraints, and discoveries that emerged from the conversation. "We decided JWT over session tokens, refresh via rotation, middleware in /api/auth/*" — that's what the implementing agent needs.

I've been building Recall — an open-source MCP memory server that provides exactly this layer. It's not Anthropic-specific, which actually makes it more flexible:

• Cross-client memory: Facts stored from any MCP client (Claude Code, VS Code Copilot, Cursor) are available to any other. Your claude.ai → Claude Code workflow becomes: store decisions during the architecture chat, recall them during implementation.
• Semantic search: recall("auth system design decisions") returns the relevant facts without needing to know which conversation they came from.
• No infrastructure: Runs locally via uvx ai-recallworks, stores facts as flat files on disk. MIT licensed.

Your Example 1 would work like this:

1. During claude.ai architecture chat, agent does remember("auth: JWT tokens, refresh via rotation, middleware at /api/auth/*, no session cookies")
2. In Claude Code, agent does recall("auth architecture decisions") → gets back the exact constraints
3. Implementation starts from the right place, no re-explanation needed.

Not a replacement for native cross-product sync (which Anthropic should build), but it works today and is provider-agnostic — same memory backend works across Claude, GPT, Gemini, whatever your team uses.

+1

+1

+ 1
I was about to make Claude code do this by some wacky automation of pulling the chat scripts from browser (with Obsidian extension) and then feeding it as context etc. Ideally this should be part of CC though with it being able to look up the convo right away. Would be perfect

I’m a solo engineering consultant, and I use claude.ai for document review and judgment work (checking report language, standards/regulatory questions, reasoning through decisions) and Claude Code for direct work on the actual project files. Both are meant to draw from the same shared context, but there’s no way for real thinking that happened in a claude.ai conversation to reach a Claude Code session without manually re-explaining or copy-pasting it over.

Even read-only, on-request access (“pull relevant context from a claude.ai conversation into this session”) would remove a lot of repeated, avoidable friction — and this use case applies to any professional using claude.ai for review/thinking and Claude Code for the underlying files, not just software developers.

+1 to this request.