Claude Session Viewer

A local web interface for browsing Claude Code session history and managing related local Claude tooling.

Quick Start

Double-click start.bat, or start the server manually:

cd .
python server.py

Then open http://localhost:8080 in your browser.

Features

• Browse Claude Code projects and session lists.
• View full conversation content, including user messages and assistant replies.
• Use local JSON caching so unchanged sessions load from cache quickly.
• Copy claude -r <session-id> recovery commands.
• Manage local MCP servers, skills, agents, and slash commands from the homepage catalog.
• Generate launch and recovery commands with permission mode, working directory, model, and prompt options.
• Scan local MCP, skill, agent, and slash command files at startup and persist the index to data/catalog.json.
• Import MCP JSON snippets and generate claude mcp add-json commands.
• Search public skill repositories and generate skill installation commands.
• Create project-level or user-level Claude agent Markdown files.
• Edit agent name, description, model, tools, and prompt from the agent detail panel.
• Create, edit, delete, enable, disable, and run agent-bound scheduled tasks.
• Discover existing Windows scheduled tasks and daily plan files for the selected agent.
• Configure QQ push profiles and agent-bound outbound notification workflows.
• Search and filter sessions.
• Follow an Anthropic-inspired visual style with a cream canvas and coral accent system.

Tech Stack

• Python standard library http.server; no required third-party backend dependency.
• Plain HTML, CSS, and JavaScript; no build step.
• Local JSON files for cache, catalog, task, connection, run, and prompt state.

Cache Model

On first load, the server parses Claude Code JSONL session files and writes normalized JSON cache files into cache/. On later loads, it compares each session file fingerprint by size and modification time:

• Unchanged files return directly from cache.
• Changed files are parsed again and the cache is refreshed.

Local Data

When server.py starts, it scans the current working directory and ~/.claude, then writes the local tool index to data/catalog.json. Homepage forms and agent workflows also persist local state under data/, including MCP import records, skill install records, prompt settings, QQ push profiles, agent tasks, agent connections, and run history.

data/ is machine-local state and is ignored by Git.

Sensitive QQ push values such as API keys and bot tokens are stored only in data/qq_push_config.json. Windows scheduled task commands reference profile names instead of embedding secrets directly in schtasks commands.

Agent Automation

Claude agent Markdown files remain the source of truth:

• Project agents: <projectRoot>/.claude/agents/*.md
• User agents: %USERPROFILE%\.claude\agents\*.md

The viewer stores automation metadata in local JSON files keyed by agent identity. Selecting an agent opens an agent detail workspace where you can inspect or modify:

• Agent prompt and frontmatter settings.
• Local cron-style task records.
• Existing Windows scheduled tasks related to the project.
• Daily plan files and recent run metadata.
• Agent-bound outbound connection settings.

Keyboard Shortcuts

• / focuses the session search field.
• Esc returns to the welcome page.
• Use claude -r <session-id> in a terminal to resume a session.

Git

This directory is a Git repository with main as the default branch. .gitignore excludes local caches, generated launcher artifacts, Python bytecode, and common build outputs.