# AgentBisect MCP server

> Enables AI coding agents to perform binary search debugging and workspace forking to pinpoint and recover from errors in multi-step trajectories.

- Repository: https://github.com/test-bit-sys/agentbisect
- Page: https://hostedmcp.org/servers/test-bit-sys/agentbisect
- Transport: —
- Stars: 3
- License: MIT

## Tools (5)

- `agentbisect_snapshot`
- `agentbisect_bisect`
- `agentbisect_rewind`
- `agentbisect_get_diff`
- `agentbisect_status`

## FAQ

### How does AgentBisect find the exact step where an agent failed?

It uses a binary search trajectory engine that locates the first broken step in O(log N) test executions.

### Does using AgentBisect affect my existing Git repository?

No, it uses a non-polluting shadow Git ledger that snapshots checkpoints inside a specific directory without affecting your real .git log.

### Can my AI agent use AgentBisect to fix itself?

Yes, agents running in environments like Cursor or Claude Desktop can use the MCP server to autonomously debug and rewind their own sessions.

### What happens after a regression is found?

You can fork the workspace to the last verified good step, which generates an intelligent prompt to help the agent resume without repeating the error.