Claude Replay
Checkpoint, search, tag, and export Claude Code sessions so you can resume long agent runs after crashes or context limits.
Overview
Claude Replay is an MCP server for the Operate phase that checkpoints Claude Code sessions so you can resume, search, tag, and export recorded conversations.
What is this MCP server?
- stdio MCP via uvx claude-replay 0.4.0 with mcp subcommand
- Session checkpoint and recovery tailored to Claude Code workflows
- Search and tag recorded sessions for finding past decisions
- Export sessions for backups, audits, or handoff notes
- Shared CLAUDE_REPLAY_DB path (~/.claude-replay/sessions.db default) across HTTP and stdio
- Version 0.4.0
- Default database: ~/.claude-replay/sessions.db
- stdio entry uses runtime argument mcp with uvx
Community signal: 1 GitHub stars.
What problem does it solve?
Long Claude Code sessions vanish or fragment when context resets, making it hard to recover decisions or continue exactly where you stopped.
Who is it for?
Solo builders running lengthy Claude Code refactors or support fixes who need searchable session history and resume points.
Skip if: Teams that only need ephemeral chats with no record, or production observability for live user traffic (use APM/error tools instead).
What do I get? / Deliverables
After install, sessions persist in SQLite-backed storage you can search, tag, resume through MCP, and export for documentation.
- Searchable tagged archive of Claude Code MCP sessions
- Resume checkpoints after disconnects or new sessions
- Exportable session records for notes, compliance, or debugging agent behavior
Recommended MCP Servers
Journey fit
Session recovery and replay support ongoing product work after shipping—when you iterate fixes and need continuity from prior agent conversations. Resume, search, and export map to iteration hygiene on recorded sessions, not greenfield frontend or launch distribution.
How it compares
Claude Code session archive MCP, not application error monitoring or git history.
Common Questions / FAQ
Who is Claude Replay for?
Claude Code users who want durable session checkpoints, search, tagging, and export instead of losing long agent threads.
When should I use Claude Replay?
Use it during Operate and iterate loops when sessions are long, interrupted, or when you must find and resume a tagged conversation.
How do I add Claude Replay to my agent?
Add the uvx claude-replay package with the mcp positional argument for stdio, and set CLAUDE_REPLAY_DB if you need a custom or shared database path.