Cursor Chat Bulk Export

Export your Cursor chat, Composer, and Agent conversations to clean Markdown files — locally on your machine, read-only, no network uploads.

Works in Visual Studio Code and Cursor (VS Code–compatible).

Author: Anas Abbas · Website: anas-code.com · Publisher: AnasAbbasCode

Why this extension?

Cursor stores conversations in local SQLite databases under your user profile. There is no built-in “export all chats to Markdown” for a project. This extension:

• Finds conversations tied to your current workspace
• Lets you pick which chats to export
• Writes one .md file per conversation plus an INDEX.md summary
• Offers a clean export (no tool-call noise, no empty “thinking” stubs) or a full raw archive

Features

Requirements

• VS Code ^1.85.0 or Cursor (compatible host)
• Windows / macOS / Linux — Cursor (or VS Code) with existing local chat data
• For very large global databases (~2 GB+ on Windows): SQLite CLI on PATH, or sqlite3.exe beside the extension out/ folder (see Large databases)

Important — do not use v0.1.0

Version 0.1.0 has known critical issues (broken discovery on newer Cursor installs, unreliable exports). Do not install or stay on 0.1.0.

If you already have 0.1.0 installed

1. Extensions → Cursor Chat Bulk Export → ⋯ → Install Specific Version… → choose 0.1.2 (or latest).
2. Or download the .vsix from the Open VSX download page → Install from VSIX… → reload the window.

In Cursor: the extension detail page may still show 0.1.0 as “current” while a newer version is already listed under Install Specific Version — that is a known Cursor marketplace cache delay, not a sign that the fix is unpublished.

The publisher may remove 0.1.0 from Open VSX so new installs cannot select it. Users who already downloaded it must update manually (steps above).

Installation

From VS Code Marketplace / Open VSX

1. Open Extensions (Ctrl+Shift+X / Cmd+Shift+X)
2. Search for Cursor Chat Bulk Export
3. Click Install
4. Reload the window if prompted

From a .vsix file

1. Download or build cursor-chat-bulk-export-0.1.0.vsix
2. In VS Code/Cursor: Extensions → ... → Install from VSIX...
3. Select the file and reload

Build from source (developers)

git clone https://github.com/anasabbasdev/cursor-chat-bulk-export.git
cd cursor-chat-bulk-export
npm install
npm run compile

Press F5 to open an Extension Development Host with the extension loaded.

Package a VSIX:

npx @vscode/vsce package

Quick start

1. Open the workspace folder whose chats you want (the same folder you use in Cursor).
2. Ctrl+Shift+P / Cmd+Shift+P → Cursor Chat Bulk Export: Export Current Workspace Chats
3. Wait for Scanning Cursor storage… (usually seconds; headers only).
4. Select conversations (Space to toggle, Enter to confirm).
5. Choose Clean export (recommended) or Full raw export.
6. Wait for Loading chat messages… (only for selected chats; large chats can take 1–3 minutes each).
7. Files appear in .cursor-chat-export/ in your project root.

Open the folder: Cursor Chat Bulk Export: Open Export Folder.

Commands

All commands are available from the Command Palette.

Export workflow (scan → pick → load → write)

Scan (fast)     →  QuickPick      →  Load messages   →  Write .md
metadata only      your selection     selected only      + INDEX.md

• Scan does not load every bubble in a 2 GB database — only titles, IDs, and estimated message counts.
• Load runs per selected conversation and fetches only bubbles referenced in composerData headers (not every historical bubbleId row).

Progress messages:

• Scanning Cursor storage…
• Loading chat messages… 3/10: <chat title>
• Exporting conversations…

Clean export vs full raw export

Clean export (recommended, default)

Produces a compact, LLM-friendly transcript:

• User and assistant text kept
• Tool-only lines removed (e.g. `tool_…`, `call_…`)
• Thinking-only placeholders removed (**Thinking**, Thinking..., etc.)
• Empty messages skipped

Good for backups, documentation, or pasting context into a new chat.

Full raw export

Includes tool calls, thinking blocks, and empty messages — useful for debugging or complete archival.

Settings

Settings JSON example:

{
  "cursorChatExport.includePossibleWorkspaceMatches": true,
  "cursorChatExport.cursorUserDataPath": ""
}

Using a copied Cursor folder (diagnostics / migration)

If you copied %APPDATA%\Cursor\User from another PC (e.g. D:\Cursor\User) only for inspection, set:

"cursorChatExport.cursorUserDataPath": "D:\\Cursor\\User"

Or set environment variable CURSOR_APPDATA to the parent of User (the folder that contains User\globalStorage and User\workspaceStorage).

Normal export should use the live Cursor data on the machine where you run Cursor, not a stale copy — otherwise chats won’t match what you see in the UI.

Expected layout:

User/
├── globalStorage/
│   └── state.vscdb          ← main DB (composerData, bubbleId); often 1–3+ GB
└── workspaceStorage/
    └── <hash>/
        ├── workspace.json
        └── state.vscdb      ← per-workspace UI / panel metadata

Output format

.cursor-chat-export/
├── INDEX.md
├── diagnostics/
│   └── conversation-discovery-report.md   ← after Diagnose Conversation Discovery
├── 2026-05-24__fix-login-session-error.md
├── 2026-05-24__docker-erp-setup.md
└── 2026-05-24__untitled-chat-03.md

Per-conversation file:

# Fix login session error

- **Workspace:** C:\Projects\MyApp
- **Exported at:** 2026-05-24T19:00:00.000Z
- **Original date:** 2026-05-23T14:32:11.000Z
- **Session type:** composer
- **Messages:** 28
- **Filtered internal messages:** 64
- **Source:** Cursor local chat storage

---

## User

How do I fix the session expiry bug in auth.ts?

---

## Assistant

The issue is in the `refreshToken` function…

INDEX.md lists each file with exported vs filtered counts and role breakdowns.

How discovery works (short)

Cursor splits data across:

1. globalStorage/state.vscdb — table cursorDiskKV: composerData:<id> (metadata + message headers), bubbleId:<composerId>:<bubbleId> (message text).
2. workspaceStorage/<hash>/state.vscdb — table ItemTable: UI state, including workbench.panel.composerChatViewPane.<tabId> whose JSON points to workbench.panel.aichat.view.<composerId>.

The extension:

1. Resolves workspace storage hashes for your folder path (and optional folder-name match).
2. Collects composer IDs from workspace panel keys.
3. Loads matching composerData from global storage (SQL scope + direct ID lookup).
4. On export, loads only bubbles listed in each composer’s headers.

Large databases (over ~2 GB)

globalStorage/state.vscdb can exceed 2 GB. Built-in sql.js cannot open files that large; the extension switches to the sqlite3 command-line tool automatically.

Windows setup

1. Install SQLite (one of):
   • winget install SQLite.SQLite
   • sqlite-tools-win-x64.zip → extract sqlite3.exe
2. Either add sqlite3 to PATH, or copy sqlite3.exe next to the extension bundle:
   • Development: MDCursorExporter/out/sqlite3.exe
   • Installed: %USERPROFILE%\.cursor\extensions\anasabbascode.cursor-chat-bulk-export-<version>\out\sqlite3.exe
   • Same pattern under %USERPROFILE%\.vscode\extensions\ if installed in VS Code

Confirm in Output → Cursor Chat Bulk Export: Opened DB via cli.

Troubleshooting

Output channel

View → Output → Cursor Chat Bulk Export. Run any command; logs appear here.

UI shows more chats than export finds

Run Diagnose Conversation Discovery. Optionally enter how many chats Cursor’s UI shows (e.g. 48).

Report path: .cursor-chat-export/diagnostics/conversation-discovery-report.md

Raw Discovery Mode shows each candidate key and whether it parsed.

Chat stuck on “Loading chat” in Cursor (never finishes)

That usually means no full local copy of the conversation (cloud/synced or never downloaded). The extension cannot export chats that Cursor itself cannot open. Titles may still appear in the list from workspace metadata.

Try: same account, internet, open the chat once on the old machine, wait for sync, then export again.

Export is slow

Expected for large histories:

• Scan: seconds (metadata only).
• Load: depends on selected chats; a chat with hundreds of messages may take 1–3+ minutes on a 2 GB DB.
• Write Markdown: usually seconds.

Export fewer chats at a time if needed.

No conversations found

1. Run Diagnose Current Workspace Chat Schema and Diagnose Conversation Discovery.
2. Ensure the correct workspace folder is open.
3. On large DBs, install sqlite3 (see above).
4. Set cursorChatExport.includePossibleWorkspaceMatches to true if the project path moved.

Stack overflow / crash on diagnose (old versions)

Current builds use a light scan on huge DBs (counts + samples, not full key enumeration). Update to the latest VSIX if you still see Maximum call stack size exceeded.

Architecture

src/
├── extension.ts                 Activate / register commands
├── logger.ts                    Output channel logger
├── types.ts                     Conversation, ExportOptions, etc.
├── storage/
│   ├── cursorStorage.ts         Resolve Cursor User path (APPDATA / settings)
│   ├── cursorDiskKV.ts          composerData + bubbleId (header-scoped load)
│   ├── workspaceScanner.ts      workspaceStorage scan + primary entry selection
│   ├── sqliteReader.ts          sql.js (WASM) for smaller DBs
│   ├── sqliteCliReader.ts       sqlite3 CLI for DBs ≥ ~1.85 GB
│   └── dbBackend.ts             Unified DB backend (sqljs | cli)
├── workspace/
│   └── workspaceMatch.ts        Path / hash / folder-name matching
├── chat/
│   ├── chatParser.ts            composerData + bubbles → Conversation
│   ├── itemTableDiscovery.ts    Panel keys + ChatSessionStore index
│   ├── schemaDiscovery.ts       ItemTable chat key heuristics
│   └── legacyParser.ts          Older ItemTable-only format
├── discovery/
│   ├── conversationDiscovery.ts Scan, defer load, hydrate for export
│   ├── discoveryReport.ts       Markdown diagnostic report
│   ├── lightDbScan.ts           Safe large-DB sampling
│   └── dedupeConversations.ts   Merge duplicate IDs
├── export/
│   ├── exportFilter.ts          Clean vs raw filtering
│   ├── markdownExporter.ts      Conversation → .md
│   ├── filenameSanitizer.ts     Safe Windows filenames
│   └── indexGenerator.ts        INDEX.md
└── ui/
    └── commands.ts                Command Palette + QuickPick pipeline

Known limitations

• Undocumented Cursor schema — may change between Cursor versions; use Diagnose commands after upgrades.
• Cloud-only / NAL chats — composerData may exist with isNAL: true while bubbles are missing locally; export shows placeholders or skips empty chats.
• Chats that never load in Cursor — not recoverable from local SQLite alone.
• Performance — very large state.vscdb + many long chats: export can take tens of minutes; select subsets.
• Workspace association — matching uses path, hash, and folder name heuristics; not 100% identical to Cursor’s UI in every edge case.

Privacy & safety

• Read-only access to Cursor databases (sql.js or sqlite3 -readonly).
• No network calls; nothing is sent to third parties.
• No writes to Cursor storage; export files go only under your project’s .cursor-chat-export/.
• Existing .md files are not overwritten — a numeric suffix is added (name-2.md).

Contributing & support

• Website: anas-code.com
• Repository: github.com/anasabbasdev/cursor-chat-bulk-export
• Issues: github.com/anasabbasdev/cursor-chat-bulk-export/issues

License

MIT — Copyright (c) 2026 Anas Abbas