Claude Conversation Browser

A small toolkit for importing your Claude conversation export into a SQLite database and browsing it from the terminal.

Files

File	Purpose
probe_schema.py	Inspect the structure of a raw export file
probe_tools.py	Deep-dive into tool use / artifact content in an export
extract.py	Import the export into SQLite and extract artifacts to disk
browse_claude.py	Terminal UI for browsing conversations and artifacts

---

Quick start

# 1. Import your export
python3 extract.py conversations.json

# 2. Browse
python3 browse_claude.py

Both scripts default to claude.db and ./artifacts in the current directory. Run them from the same directory as your export file for the simplest setup.

---

Import

python3 extract.py conversations.json [--db claude.db] [--artifacts ./artifacts]

Reads conversations.json (the file Anthropic emails you as a ZIP attachment when you request an export from Settings → Account → Export Data).

What it does:

• Populates claude.db with conversations, messages, content blocks, attachments, and file references.
• Reconstructs the final state of each file artifact by replaying create_file and str_replace tool calls in order. Files written via bash heredocs (cat > file << 'EOF') are also captured.
• Writes artifact files to artifacts/<conv-uuid-prefix>/<conv-slug>/filename.
• Records each artifact's local disk path in the DB so the browser can show it and you can cp it directly from the shell.
• Skips conversations where no message has any content (these are empty shells that Anthropic sometimes includes in exports).
• Safe to re-run: uses INSERT OR REPLACE throughout and overwrites artifact files on disk.

Artifact layout:

artifacts/
  03cba4fd/                        ← first 8 chars of conversation UUID
    Dealing-range-builder/
      dr_builder.py
      test_dr_builder.py
  019e1e0f/
    Weather-Overground/
      main.go
      go.mod

Known limitation: if a file was modified by a bash command (e.g. go mod tidy, sed -i) between a create_file and a subsequent str_replace, the in-memory state can drift and the patch may not apply. The importer prints a warning when this happens:

[warn] str_replace on /path/to/file failed: old_str not found in content

The artifact is still written with whatever content was reconstructed up to that point.

---

Browse

python3 browse_claude.py [--db claude.db]

Requires no dependencies beyond the Python standard library. Works in any terminal that supports curses (tested on Alacritty). Mouse wheel scrolling is supported.

Conversation list

Key	Action
j / k or arrows	Move selection
Enter	Open conversation
1–9	Jump directly to nth conversation
s	Cycle sort order: date → name → messages → artifacts
/	Search conversation titles; n / N next / prev match
Ctrl-U / Ctrl-D	Scroll half-page
q	Quit

Conversation view

The thread is rendered as a continuous scrollable document. Each block (human turn, assistant turn, tool call, thinking) has a coloured left gutter bar:

Colour	Sender
Cyan	Human
Green	Assistant
Yellow	Tool / artifact
Magenta	Thinking

create_file and bash heredoc artifacts are rendered as framed code boxes showing the first 30 lines. The artifact's local disk path is shown at the bottom-right of the box.

Key	Action
j / k or arrows	Scroll one line
Ctrl-U / Ctrl-D	Scroll half-page
Mouse wheel	Scroll
m / M	Jump to next / prev block
t	Toggle thinking blocks visible in the viewport
a	Toggle artifact panel
1–9 (artifact panel open)	Open nth artifact in $PAGER
c	Copy first artifact path to clipboard
e	Export conversation to Markdown
/	Search; n / N next / prev match
q / Esc	Back to conversation list

Exporting to Markdown writes a file named conversation_<uuid-prefix>.md in the current directory containing the plain text of all messages (tool calls omitted).

Copying artifact paths: c copies the local disk path to the system clipboard (via pbcopy, xclip, xsel, or wl-copy, whichever is available). If none are found it prints the path to the status bar instead so you can triple-click and copy manually. Once you have the path:

cp /path/to/artifacts/03cba4fd/Dealing-range-builder/dr_builder.py ~/myproject/

---

Probes

The probe scripts were used to reverse-engineer the export schema before writing the importer. You won't normally need them, but they're useful if Anthropic changes the export format or you want to inspect a new export before importing.

# General schema overview
python3 probe_schema.py conversations.json

# Tool use / artifact deep-dive
python3 probe_tools.py conversations.json

Both are read-only and write nothing to disk.

---

Database schema

conversations   uuid, name, summary, account_uuid, created_at, updated_at

messages        uuid, conversation_uuid, parent_message_uuid,
                sender, text, created_at, updated_at

content_blocks  id, message_uuid, type,
                start_timestamp, stop_timestamp,
                text, citations,                     -- type=text
                thinking, summaries, signature,       -- type=thinking
                truncated, cut_off,
                tool_id, tool_name, tool_input,       -- type=tool_use
                integration_name, is_mcp_app,
                mcp_server_url,
                tool_use_id, result_name,             -- type=tool_result
                result_content, is_error

attachments     id, message_uuid, file_name, file_type,
                file_size, extracted_content

files           id, message_uuid, file_uuid, file_name

artifacts       id, conversation_uuid, message_uuid,
                original_path, disk_path, file_name, created_at

You can query the database directly with any SQLite client. Example:

# List all artifacts across all conversations
sqlite3 claude.db "
  SELECT c.name, a.file_name, a.disk_path
  FROM artifacts a
  JOIN conversations c ON c.uuid = a.conversation_uuid
  ORDER BY a.created_at DESC;
"

# Full text search (basic)
sqlite3 claude.db "
  SELECT c.name, m.sender, substr(m.text, 1, 120)
  FROM messages m
  JOIN conversations c ON c.uuid = m.conversation_uuid
  WHERE m.text LIKE '%dealing range%'
  ORDER BY m.created_at;
"