agarwalvishal/claude-chat-exporter

Claude Chat Exporter – What it Is

A tiny, open‑source browser script that lets you download any conversation you have on Claude.ai as a clean Markdown file. It runs completely in your browser (no server), pulls the raw markdown Claude generated directly from Claude’s private API, and saves the result locally.

Why It Exists

• The web UI only shows a small window of a long chat and relies on fragile CSS selectors for copy‑buttons.
• Exporting via the UI can miss artifacts, files, charts, or attachments.
• For knowledge‑base tools like Obsidian or Retrieval‑Augmented Generation (RAG) pipelines you need a well‑structured, metadata‑rich markdown document.

Core Features

Feature	What You Get
One‑click bookmarklet	Drag a button to your bookmarks bar, open a Claude chat, click it → a .md file downloads automatically.
Perfect markdown fidelity	Reads Claude’s source markdown from the API – no HTML conversion, so tables, math, code, etc. are byte‑perfect.
Full element support	Tables, LaTeX math, code blocks, nested lists, links, headings, artifacts, created files, charts/diagrams (as fenced code), and all attachments.
Obsidian‑ & RAG‑ready output	YAML front‑matter (title, source, model, exported) plus a top‑level # Human / # Claude heading for each turn – ideal for note‑taking or chunking for retrieval.
Timestamps	Per‑message timestamps from Claude’s API.
Smart filenames	Uses the conversation title from the API (fallback claude_conversation.md).
Privacy‑first	No backend, no tracking; the only network request is the single authenticated call to Claude’s own API.
Robust to UI changes	Works off a stable data contract rather than CSS selectors.

How It Works (in a nutshell)

1. Fetch – Calls GET /api/organizations/{orgId}/chat_conversations/{conversationId} (same‑origin, using your session cookie). The response contains the conversation title, model, and every message.
2. Order – Reconstructs the current branch of the conversation tree by walking from the current_leaf_message_uuid up through parent_message_uuid links, then reversing the list.
3. Extract – Iterates each message’s content blocks in order, emitting raw text blocks and rendering special tool_use blocks (artifacts, create_file, visualize widgets) as labelled fenced code.
4. Build Markdown – Prepends YAML front‑matter, adds a # Human — … / # Claude — … header for each turn, inserts timestamps, and places attachments (images embedded, text files block‑quoted, other docs linked).
5. Download – Triggers a browser download of the assembled .md file.

Using It

• Bookmarklet – Visit the install page, drag the “📥 Claude Export” button to your bookmarks bar, open a Claude chat, click the bookmark. The script updates itself automatically.
• Console mode – Copy the claude-chat-exporter.js source, paste it into the browser dev console on a Claude conversation page, and press Enter. The script shows a tiny status box and then downloads the file.

Privacy & Security

• Runs entirely client‑side; nothing is sent to the author’s servers.
• The only network request is the authenticated GET to Claude’s own backend (the same request the web UI already makes).
• No data is stored after the download; the bookmarklet’s update check only fetches public script files.

Limitations

• Works only on the Claude web UI while you are logged in.
• Depends on Claude’s undocumented internal API – a breaking change would require a script update.
• Does not export binary attachment data (images/PDFs are linked, not bundled) and skips some tool calls (web search, bash, etc.).
• Requires JavaScript enabled in the browser.

Who Might Use It

• Researchers or developers who want a reproducible record of their Claude sessions.
• Knowledge‑base builders using Obsidian, Notion, or similar tools.
• Anyone building a RAG pipeline that needs clean, timestamped markdown with metadata.

---

The project is MIT‑licensed and maintained by the author; sponsorship helps keep it up‑to‑date with Claude’s evolving API.