Skip to main content
MCP server CX Codex

orcareplay — Claude Code MCP server: replay and setup

Records coding-agent runs, replays them offline, and lets you fork a run onto another model. Useful when you need to inspect a failure.

Install into Claude Code · Platform CLI All methods and platforms ↓
claude mcp add orcareplay -- npx -y orcareplay
Open on GitHub

What orcareplay does

orcareplay records a coding-agent session through a local proxy, then replays the saved trace without making another model call. You can inspect which tool call changed a file, rerun the same session after the terminal is closed, or branch from a chosen step and use a different model.

The server runs through npx and does not require edits to the agent. The proxy observes requests, responses, tool calls, and tool results through environment variables. During a normal replay, requests that are missing from the trace are blocked, while recorded tool calls still run for real. A command such as curl, or an MCP server that opens its own connection, sits outside that guarantee. Node.js 20+ is required.

What it helps with

  • record a coding-agent session
  • replay a failure without another model call
  • identify the tool call that changed a file
  • fork a run from a selected step
  • compare another model on the same trace

Example prompts for the agent

  • Show me the last recorded run and identify the step that removed the file.
  • Replay the latest run offline without contacting a model.
  • Fork the recording from step 4 onto another model and compare the result.

How to install orcareplay

Platform CLI

claude mcp add orcareplay -- npx -y orcareplay

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about orcareplay

How do I install orcareplay in Claude Code?

Run claude mcp add orcareplay -- npx -y orcareplay. This registers the orcareplay MCP server with Claude Code.

What does orcareplay do?

It records coding-agent sessions, replays them from a saved trace without calling a model, and lets you continue from a chosen step with another model.

What happens to network calls during replay?

A normal replay blocks requests that are not present in the trace. Recorded tool calls still execute, while tools that open their own network connections are outside the replay guarantee.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor